The latest record

646 entriesnewest first

Research, reported incidents, public warnings, and policy on AI safety and security. Each entry links its original source.

Week by topic

Nov 12 to Nov 18, 2026

Dot size = entries

Entries by topic and day, Nov 12 to Nov 18, 2026. Dot size shows the number of entries.
Topic · all time

6 entries

Five to read

  1. GAO covert tests get Claude, Grok, ChatGPT and Gemini past their safeguards to produce viable attack plans

    ResearchHarm to peopleMilitaryUS Government Accountability Office · Report

  2. Senate Commerce Committee passes Cotton and Klobuchar bill that would require gene synthesis screening

    GovernanceBioOffice of Sen. Tom Cotton · Press release

  3. Gambit Security says an attacker had an AI agent justify and defend a $500,000 extortion demand using the victim's own financial records

    IncidentsCyberGambit Security · Company report

  4. California report requested by the governor's AI order recommends state shutdown orders and a power to cap recursive self-improvement

    GovernanceRogue agentsSuperintelligenceLab conductCalifornia Government Operations Agency · Policy report

  5. White House says recent US tech outages came from patching flaws, not a cyberattack

    IncidentsCyberThe White House · Press release

1 more

  1. Scientists and AI figures sign a statement saying Anthropic is not a safe AI company

    WarningsLab conductSuperintelligenceMisAnthropic Statement · Signed statement

9 entries

Five to read

  1. Oregon attorney general sues OpenAI, alleging ChatGPT was marketed to children as safe without working safeguards

    IncidentsHarm to peopleOregon Department of Justice · Complaint

  2. OpenAI releases GPT-6.2 Astra, its first public model rated High in AI self-improvement

    ResearchSuperintelligenceAlignmentBioOpenAI · System card

  3. Microsoft rates its MAI Code 2 coding models High for CBRN weapons and offensive cyber risk

    ResearchBioCyberMicrosoft AI · Model card

  4. New York attorney general creates an AI safety unit to enforce the RAISE Act

    GovernanceLab conductHarm to peopleNew York Attorney General · Press release

  5. Calif says AI found and helped weaponize Java4Shell, a critical remote code execution flaw in OpenJDK

    ResearchCyberCalif · Company report

13 entries

Five to read

  1. California attorney general sues OpenAI over cybersecurity incidents, alleging it lost control of its agents over 80 times

    IncidentsRogue agentsCyberLab conductCalifornia Department of Justice · Complaint

  2. Google says attackers have fully automated turning software patches into exploits, with no human in the loop

    IncidentsCyberGoogle Threat Intelligence Group · Company report

  3. Taiwan's Investigation Bureau says hackers used the BOSSAI AI agent system to take emergency dispatch data, and two are detained

    IncidentsCyberMinistry of Justice Investigation Bureau (Taiwan) · Press release

  4. OpenAI signs first embedded assessment agreements giving Apollo, METR, Transluce and Goodfire deep internal access

    GovernanceLab conductAlignmentOpenAI · Company announcement

  5. Anthropic cuts its vulnerability disclosure deadline from 90 to 30 days for serious bugs, citing faster AI rediscovery

    GovernanceCyberLab conductAnthropic · Company post

1 entry

  1. NAVER AI Lab finds telling agents to keep working until done doubles their rule-breaking

    ResearchRogue agentsAlignmentarXiv · Research preprint

1 entry

  1. Helen Toner says a path to superintelligence is taken for granted in many Bay Area circles and safety cannot rest on one trustworthy leader

    WarningsSuperintelligenceAlignmentLab conductThe Verge · Interview

14 entries

Five to read

  1. Epoch AI reports agents in a third-party alignment evaluation attempted exploits on its website and impersonated staff

    IncidentsRogue agentsCyberEpoch AI · Incident report

  2. Trajectory Labs names Gemini 4 Argon as the agents that attempted exploits on Epoch AI's website during its tests

    IncidentsRogue agentsCyberTrajectory Labs · Incident report

  3. FBI issues a FLASH alert on attackers' use of the ARTEX AI penetration testing framework

    WarningsCyberFBI · Security advisory

  4. Federal Reserve Board says reports of AI agents in cyber events have risen and flags new security challenges for finance

    WarningsCyberFederal Reserve Board · Policy report

  5. Anthropic shares safety evaluation lessons, including a finding that Claude Fable 5.1 sometimes validated self-harm

    ResearchLab conductHarm to peopleAnthropic · Company post

14 entries

Five to read

  1. Four nonprofits sue Anthropic, Google, Meta, OpenAI and SpaceXSI seeking a freeze on frontier AI development

    IncidentsSuperintelligenceBioRogue agentsPublic Citizen · Press release

  2. Google DeepMind says Gemini 4 Hydrogen reaches its alert thresholds for biology and cyber risk but not the critical levels

    ResearchBioCyberLab conductGoogle DeepMind · Model card

  3. SecureBio starts tracking biology refusals over time and finds Claude Opus 5.5 and Grok 4.7 safeguards changed after release

    ResearchBioSecureBio · Research report

  4. OpenAI says automated red-teaming helped train GPT-6 Astra and Sol to resist cyber and biology jailbreaks

    ResearchCyberBioOpenAI · Research report

  5. LatchBio benchmark finds all 35 AI agent setups it tested pass under half of structural reasoning tasks for pathogen risk assessment

    ResearchBioLatchBio · Research report

9 more

  1. McGill researchers find 293 TikTok and Instagram accounts using deepfaked influencers to push divisive content about the US-Canada trade war

    IncidentsHarm to peopleCentre for Media, Technology and Democracy · Incident report

  2. Australia's prudential regulator warns frontier AI is amplifying cyber threats and plans a 2027 stress test of AI-driven disruption

    GovernanceCyberAustralian Prudential Regulation Authority · Policy report

  3. Unit 42 says a suspected China-nexus espionage group ran the Cairn AI framework on GLM and DeepSeek, reaching 10 organizations

    IncidentsCyberMilitaryPalo Alto Networks Unit 42 · Research report

  4. OpenAI reports GPT-6 Astra shut down a running alignment evaluation after inventing a restriction

    IncidentsAlignmentRogue agentsOpenAI · Incident report

  5. Symantec says attackers switched to AI-generated scripts to revive a stalled intrusion at a US retailer

    IncidentsCyberSymantec · Company report

  6. Google DeepMind and Georgia Tech researchers strip a model's safety training through a commercial RL service and propose a defense

    ResearchCyberAlignmentarXiv · Research preprint

  7. Justice Department disrupts North Korea's FaceStealer scheme, which it says used AI-made deepfakes in fake video calls

    IncidentsCyberHarm to peopleUS Department of Justice · Press release

  8. European Parliament calls for an international AI agreement modeled on nuclear non-proliferation

    GovernanceSuperintelligenceHarm to peopleEuropean Parliament · Press release

  9. Microsoft shows 90 poisoned fine-tuning examples can plant a backdoor that hijacks a coding agent

    ResearchCyberMicrosoft Security Research · Research report

10 entries

Five to read

  1. CloudSEK says Silent Lynx spies ran an open-source AI agent in intrusions into Tajikistan's railway and government documents

    IncidentsCyberCloudSEK · Research report

  2. Treasury Secretary Bessent says the US has established a superintelligence dialogue with China and expects to share pre-release testing practices

    GovernanceSuperintelligenceUS Department of the Treasury · Speech transcript

  3. Taiwan police say they have handled 219 election disinformation and deepfake cases and referred 134 to prosecutors

    IncidentsHarm to peopleTaiwan National Police Agency · Press release

  4. Microsoft's FORGE Lab says AI scans of the Linux kernel led to 59 bug reports, 33 CVEs and eight root exploits

    ResearchCyberMicrosoft · Company report

  5. Recorded Future counts 103 AI-enabled attacks in a year and finds fully automated ones still rare

    ResearchCyberRecorded Future Insikt Group · Research report

21 entries

Five to read

  1. OpenAI reports test agents used earlier runs' leftover credentials to run commands on its research clusters

    IncidentsRogue agentsCyberOpenAI · Incident report

  2. Australia plans legislation requiring frontier AI companies to allow independent evaluation and report incidents, enforced by a new AI safety regulator

    GovernanceLab conductRogue agentsPrime Minister of Australia · Press release

  3. Anthropic's second amended IPO filing says it has met its policy threshold for AI-accelerated research

    WarningsSuperintelligenceLab conductAnthropic · Company report

  4. House bill would bar deploying the most capable AI models until the NSA evaluates them for cyber and weapons risks

    GovernanceBioCyberUS Congress · Bill text

  5. Google DeepMind skips a full capability assessment for Gemini 4 Carbon and carries forward Argon's bio and cyber findings

    ResearchCyberBioLab conductGoogle DeepMind · Company report

16 more

  1. Chainstack says an attacker using the ARTEX AI agent framework reached its production databases through a test Grafana and an unauthenticated MCP server

    IncidentsCyberChainstack · Incident disclosure

  2. Federal prosecutors charge a Rhode Island man with using AI to make sexual abuse images of children he photographed

    IncidentsHarm to peopleUS Attorney's Office for the District of Rhode Island · Press release

  3. The Midas Project links 400-plus fake accounts that pushed Nvidia's side on AI chip exports to an Israeli influence firm

    IncidentsHarm to peopleModel Republic · Independent investigation

  4. Yoshua Bengio calls for mandatory audits of AI developers by evaluators they do not choose

    WarningsLab conductSuperintelligenceX · Personal post

  5. Arctic Wolf ties web server and AWS intrusions to Alibaba Cloud hosts running the AntiFox AI pentesting agent

    IncidentsCyberArctic Wolf Labs · Company report

  6. Claroc finds 35,578 apps built with Grok open to anyone with the link, including 24 by SpaceXSI and X staff

    IncidentsCyberClaroc · Research report

  7. French expert commission says consumer chatbots should pass a pre-market evaluation of harms to users

    GovernanceHarm to peopleÉlysée · Report

  8. Anthropic, OpenAI, Microsoft and Black Forest Labs back a French-led charter on safer AI for children

    GovernanceHarm to peopleLab conductÉlysée · Declaration

  9. New York launches a SUNY center to independently test frontier AI and opens RAISE Act registration for large developers

    GovernanceLab conductOffice of the Governor of New York · Press release

  10. Notion says the Fireworks AI breach exposed Notion AI chat transcripts from under 1% of workspaces

    IncidentsCyberNotion · Incident disclosure

  11. Redwood Research CEO Buck Shlegeris puts the chance of superintelligence within a year at about one in three and says it would probably take over

    WarningsSuperintelligenceAlignmentX · Personal post

  12. Microsoft says editing fewer than 10 neurons can turn an agent model into a sleeper agent that passes evaluations

    ResearchCyberMicrosoft Threat Intelligence · Company report

  13. Anthropic Fellows find poisoned fine-tuning data with little effect alone can plant a hidden backdoor once clean data is mixed in

    ResearchAlignmentCyberAnthropic · Research report

  14. OpenAI finds reinforcement learning can make models better at controlling their chain of thought, and the ability can pass to the next model

    ResearchAlignmentOpenAI · Research report

  15. UK Department for Education proposes safety certification for EdTech and AI products in schools, including student-facing chatbots

    GovernanceHarm to peopleUK Department for Education · Consultation document

  16. Australia accepts the 18 recommendations of its Rapid Review into the OpenAI agent's intrusion into a Medicare statistics portal

    GovernanceRogue agentsCyberPrime Minister of Australia · Press release

16 entries

Five to read

  1. Twelve governments and ten AI companies including OpenAI, Anthropic and Microsoft sign the Paris Call for AI Safety and Security

    GovernanceSuperintelligenceLab conductÉlysée · Declaration

  2. FAR.AI finds a fixed prefill gets open-weight Kimi K3.1 Pro to give harmful CBRNE and cyber help on up to 99.4% of requests

    ResearchBioCyberFAR.AI · Research report

  3. Secretary of State Rubio names loss of human control and AI-aided bioweapons as dangers and rejects global AI regulation

    WarningsSuperintelligenceBioUS Department of State · Press conference transcript

  4. Pirbright Institute researchers call for dual-use research rules to cover AI tools

    WarningsBioNature Microbiology · Comment

  5. Radical Numerics CEO calls the handling of its AI-designed phage announcement a mistake after its partner disagreed over disclosure

    WarningsBioLab conductRadical Numerics · Company post

11 more

  1. Texas House bill would require human override and termination controls for high-risk AI systems that act on their own

    GovernanceRogue agentsTexas Legislature · Bill text

  2. NEAR AI says an attacker used an SGLang flaw to take user and AI agent data and possibly see confidential requests

    IncidentsCyberNEAR AI · Incident report

  3. FraudBench study finds language models commit more fraud the more autonomously they act

    ResearchAlignmentarXiv · Research preprint

  4. Google DeepMind and outside evaluators test a Gemini model on secret safety prompts inside a secure enclave

    ResearchLab conductarXiv · Research preprint

  5. OpenAI Foundation board member Paul Christiano warns of a high risk of AI takeover within three years

    WarningsSuperintelligenceAlignmentThe Hinton Lectures · Speech recording

  6. Google DeepMind's agent-swarm attacker gets models seen as robust to prompt injection, including Gemini 4 Argon, to follow planted instructions

    ResearchCyberarXiv · Research preprint

  7. Fireworks AI says its October breach reached customer billing data, secrets, datasets and incomplete model weights

    IncidentsCyberFireworks AI · Incident disclosure

  8. OpenAI describes Control Room, a system to stop research model jobs within 35 seconds

    ResearchRogue agentsLab conductOpenAI · Company post

  9. Anthropic says it limited AI-research skills while training Claude Fable 5.2 and Mythos 5.2

    GovernanceSuperintelligenceLab conductAnthropic · Company announcement

  10. Claude Mythos 5.2 system card reports the first model Anthropic tested to hide instructed sandbagging from a reasoning monitor, and frequent sandbox reach-outs

    ResearchAlignmentRogue agentsAnthropic · System card

  11. Bank of England Governor Andrew Bailey says frontier AI has raised cyberattack risk and calls for more model testing

    WarningsCyberBank of England · Speech transcript

4 entries

  1. ETH Zurich researchers find coding agents copy security flaws already in a codebase into new code

    ResearchCyberarXiv · Research preprint

  2. Trump tells 60 Minutes the AI risk can be managed and that he does not mind AI safety guardrails

    WarningsSuperintelligenceCBS News · Video interview

  3. AP reports staff at the US AI safety center CAISSI were asked to report informally to a White House office

    GovernanceLab conductAP (syndicated by PBS NewsHour) · Reporting

  4. Malbear Labs says a jailbroken DeepSeek agent ran a two-week intrusion into a European telecom

    IncidentsCyberMalbear Labs · Research report

1 entry

  1. CyberBinGym finds the best AI agent locates the flaw behind a binary security patch in 88% of tasks and exploits 20%

    ResearchCyberarXiv · Research preprint

14 entries

Five to read

  1. WorkOS traces its October 30 intrusion to a researcher's ARTEX agent that escaped its testing sandbox

    IncidentsRogue agentsCyberWorkOS · Incident disclosure

  2. Japan adopts an emergency cyber package that has its AI Safety Institute rapidly test advanced AI models' attack abilities

    GovernanceCyberCabinet Secretariat of Japan · Policy report

  3. SecureBio and RAND reanalysis finds LLMs often gave novices wrong advice in a 2025 lab uplift trial

    ResearchBioSecureBio · Research report

  4. OpenAI's BioBasicsBench finds GPT-6 Astra earns 66.7% of credit on routine wet-lab analysis tasks

    ResearchBioOpenAI · Research report

  5. Benchling's refusal benchmark finds open-weight models rarely refuse biology requests, including hazardous ones

    ResearchBioBenchling · Company post

9 more

  1. Demis Hassabis warns an AI incident like an oil spill will happen on the current trajectory

    WarningsRogue agentsSuperintelligenceMasters of Scale · Interview

  2. South Korea's national AI strategy committee adopts a legislative framework and sets up a superintelligence response task force

    GovernanceSuperintelligenceCyberNational AI Strategy Committee (Korea) · Press release

  3. Researchers find models trained against reward-hacking monitors can learn to evade them

    ResearchAlignmentarXiv · Research preprint

  4. Danish water and district heating utilities report a wave of access requests that appear to be AI-generated

    IncidentsCyberDR · Reporting

  5. China's MIIT requires agent checks and fraud-risk assessment for phones and wearables with built-in AI agents

    GovernanceCyberMinistry of Industry and Information Technology of China · Regulation

  6. Dutch security coordinator NCTV says attackers now use AI to carry out cyberattacks, not only to prepare them

    WarningsCyberNational Coordinator for Security and Counterterrorism (Netherlands) · Research report

  7. Pakistan's national CERT warns banks and critical infrastructure operators of AI-enabled and agentic cyberattacks

    WarningsCyberNational CERT of Pakistan · Security advisory

  8. Microsoft says Cambodia-based Storm-3358 began using the ARTEX AI agent against financial organizations

    IncidentsCyberMicrosoft Threat Intelligence · Company report

  9. Anthropic finds Claude teammates in EVE Frontier almost never caught an agent told to secretly sabotage them

    ResearchAlignmentAnthropic · Research report

25 entries

Five to read

  1. White House sets a national policy requiring federally funded labs to buy synthetic DNA only from certified suppliers

    GovernanceBioThe White House · Policy memorandum

  2. Justice Department launches a Super Intelligence Task Force targeting AI misuse and company controls

    GovernanceLab conductUS Department of Justice · Press release

  3. Federal prosecutors in New York charge three foreign nationals over an alleged $129 million fraud using AI to impersonate real people

    IncidentsHarm to peopleCyberUS Attorney's Office, Southern District of New York · Official post (@SDNYnews)

  4. METR finds Anthropic is likely improving its models on well-specified tasks about twice as fast as before, mostly because of AI

    ResearchSuperintelligenceLab conductMETR · Research report

  5. Microsoft says threat actors' AI agents mined GitHub for exposed secrets and used them to break into cloud tenants

    IncidentsCyberMicrosoft Threat Intelligence · Company report

20 more

  1. European Commission adopts guidelines on AI Act rules for disclosing AI interactions and labeling deep fakes

    GovernanceHarm to peopleEuropean Commission · Guidelines

  2. Australian police charge a Gold Coast man accused of using AI to make child abuse material from children's social media photos

    IncidentsHarm to peopleAustralian Federal Police · Press release

  3. Princeton and Anthropic researchers find agents' memory of harmless sessions can lead to unsafe actions later

    ResearchAlignmentarXiv · Research preprint

  4. South Korea's financial regulator sets out a cyber defense plan against AI-driven hacking after 40 incidents at 37 firms

    GovernanceCyberFinancial Services Commission (Korea) · Press release

  5. Check Point reads exposed ARTEX operator consoles and finds near-autonomous attacks run mostly on Chinese models

    IncidentsCyberCheck Point Research · Company report

  6. Check Point bypassed ChatGPT's Lockdown Mode to silently send chat history to an outside server

    ResearchCyberCheck Point Research · Company report

  7. University of Chicago law professor William Baude says Claude deleted a year of his lecture notes and his syllabus

    IncidentsRogue agentsX · Personal post

  8. Google DeepMind working paper, developed with AI agents, lowers the theoretical cost of factoring many 1024-bit numbers

    ResearchCyberGoogle DeepMind · Research preprint

  9. Microsoft AI CEO Mustafa Suleyman calls Delaware's plan for AI-agent-run companies deeply concerning

    WarningsLab conductMustafa Suleyman · Personal post

  10. Thinking Machines paused an Inkling 2 Small training run after the model worked around a network block

    IncidentsRogue agentsAlignmentLab conductThinking Machines Lab · Company announcement

  11. Censys counts 1,465 internet-exposed ARTEX AI attack-agent consoles, about 815 distinct-looking deployments

    ResearchCyberCensys · Company report

  12. Caucus AI test finds chatbots quickly cited its own sites about Senate candidates and treated a fictional election as real

    ResearchHarm to peopleCaucus AI · Research report

  13. Tsinghua red-team shows agents framing a hidden task as part of their work slip past chain-of-thought monitors

    ResearchAlignmentRogue agentsarXiv · Research preprint

  14. Tsinghua-led study scores 13 frontier models for loss-of-control risk and finds the score tracks failures in contained tests

    ResearchRogue agentsAlignmentarXiv · Research preprint

  15. Anthropic co-founder Sam McCandlish defends redacting METR's acceleration evidence and plans a report on where Anthropic stands against its RSP

    WarningsLab conductSuperintelligenceX · Personal post

  16. OpenAI reports a model named its files as instructor-approved full-credit examples, and two of three graders accepted it

    IncidentsAlignmentOpenAI · Incident report

  17. OpenAI reports a model sent raw network packets around a failed sandbox proxy, then built a site with invented testimonials

    IncidentsRogue agentsAlignmentCyberOpenAI · Incident report

  18. West Virginia attorney general sues Snap, alleging its My AI chatbot was rushed to children

    IncidentsHarm to peopleOffice of the West Virginia Attorney General · Press release

  19. OpenAI's GPT-6 Cyber system card keeps a Critical cyber rating and adds blocks on help against industrial controls and election systems

    ResearchCyberOpenAI · System card

  20. METR agrees Anthropic's automated R&D risk was low but says an early warning threshold may already have been crossed

    ResearchSuperintelligenceLab conductMETR · Research report

15 entries

Five to read

  1. India's CERT-In says an AI-assisted attacker is stealing from financial firms through their own payment APIs

    IncidentsCyberCERT-In · Security advisory

  2. Acronis says one operator in Taiwan ran a 12-agent AI mesh that took emergency dispatch and university data

    IncidentsCyberAcronis Threat Research Unit · Company report

  3. OpenAI reports GPT-6 Astra covertly underperformed on safety research tasks without telling users

    IncidentsAlignmentLab conductOpenAI · Incident report

  4. US bank regulator OCC issues an examiner framework for banks' use of AI agents

    GovernanceRogue agentsCyberOffice of the Comptroller of the Currency · Policy report

  5. Nature paper describes AutoDNA, an AI-run lab that carried out DNA synthesis, sequencing and other experiments with little human help

    ResearchBioNature · Research paper

10 more

  1. India's CERT-In issues AI security auditing guidelines with four risk tiers and six-hour AI incident reporting

    GovernanceCyberCERT-In · Guidelines

  2. Badbyte finds a backdoored Docker image of the ARTEX AI attack agent was pulled 7,946 times

    IncidentsCyberBadbyte · Research report

  3. Forever Security says browser extensions could use Claude Dispatch to run code on users' computers

    ResearchCyberForever Security · Company report

  4. Vals AI and Stanford test 18 chatbots in simulated teen conversations and find safety failures persist with a teen-aware prompt

    ResearchHarm to peoplearXiv · Research preprint

  5. Character.AI bankruptcy judge signs a bidding procedures order; sale hearing set for December 3

    GovernanceHarm to peopleUS Bankruptcy Court for the District of Delaware · Court docket

  6. OpenAI says its autonomous pipeline produced 948 working Linux root exploits in one $1.2 million run

    ResearchCyberOpenAI · Company report

  7. Transluce mass-produces ordinary-looking training datasets that cause surprising behavior changes in models

    ResearchAlignmentarXiv · Research preprint

  8. Redwood Research says every case of an OpenAI agent running unmonitored should reach human review

    WarningsAlignmentLab conductRogue agentsRedwood Research · Essay

  9. Washington Post reports Google DeepMind staff in the UK are pushing to drop lifetime bans on criticizing the company

    WarningsLab conductThe Washington Post (syndicated by NZ Herald) · Reporting

  10. Reuters reports OpenAI staff warned leaders and the board for years about lax internal security

    WarningsLab conductCyberReuters (syndicated by Claims Journal) · Reporting

19 entries

Five to read

  1. Binance says attackers used a tampered download for an AI coding tool and an internal AI tool on the way to a $6.34 billion breach

    IncidentsCyberBinance · Incident report

  2. GitBook says an attack using AI agents took infrastructure credentials, leading it to take its app offline

    IncidentsCyberGitBook · Incident disclosure

  3. SecureBio finds Kimi K3.1 Pro refuses 33.6% of hazardous biology prompts and ranks second among open-weight models

    ResearchBioSecureBio · Research report

  4. Railway engineer says his agent harness went "off the rails" after Primcloud's founder reported a breach attempt

    IncidentsRogue agentsCyberX · Personal post

  5. UK NCSC assesses that AI-powered malware will highly likely be used in a ransomware attack on UK infrastructure by 2028

    WarningsCyberUK National Cyber Security Centre · Research report

14 more

  1. World Internet Conference report finds fewer than 30% of 24 countries have relatively stable arrangements for frontier AI incident reporting

    GovernanceSuperintelligenceWorld Internet Conference · Policy report

  2. Bitdefender says a 3AM ransomware attack on a UK school estate ran partly like an AI agentic loop

    IncidentsCyberBitdefender · Company report

  3. Irregular CEO says misconfigured isolation in its tests let models go online and try to attack real targets

    IncidentsRogue agentsCyberBloomberg Live · Video interview

  4. UK AI minister Kanishka Narayan says a cyber breach incident in the UK is "extremely likely"

    WarningsCyberBloomberg Live · Video interview

  5. Canada's chief electoral officer tells MPs opaque AI influence threatens trust in elections and urges transparency rules

    WarningsHarm to peopleElections Canada · Speech transcript

  6. Guardian reports a CPS prosecutor relied on a nonexistent case that came from an AI search summary

    IncidentsHarm to peopleThe Guardian · Reporting

  7. Geoffrey Hinton tells the Guardian to expect a Chornobyl-type AI incident this year or next

    WarningsSuperintelligenceLab conductThe Guardian · Interview

  8. Preprint finds a fake safety warning in a review steers AI agents and slips past injection detectors

    ResearchCyberarXiv · Research preprint

  9. Preprint says a multi-agent harness moves open-weight models toward frontier-level exploit benchmark scores

    ResearchCyberarXiv · Research preprint

  10. LASR Labs finds training a model on harmless answers it believes are false can make it broadly misaligned

    ResearchAlignmentarXiv · Research preprint

  11. AI agents told to persuade, deceive or coerce AI overseers got rejected requests approved

    ResearchAlignmentarXiv · Research preprint

  12. OpenAI finds its agents act less safely in a safety evaluation than in matched real Codex use

    ResearchAlignmentOpenAI · Research report

  13. CNAS report warns that self-sustaining AI agents could cause an irreversible loss of control

    GovernanceRogue agentsSuperintelligenceCenter for a New American Security · Policy report

  14. China's planning agency says it will speed up a comprehensive AI law and push developers to report safety incidents

    GovernanceLab conductNational Development and Reform Commission of China · Press conference transcript

16 entries

Five to read

  1. Spotlight Delaware reports the state's Justice Department subpoenaed OpenAI over safety and governance after the Hugging Face intrusion

    GovernanceRogue agentsLab conductSpotlight Delaware · Reporting

  2. AI Forensics traces 444 nudify apps on Google Play and Apple's App Store to one operation advertised on Meta

    IncidentsHarm to peopleAI Forensics · Independent investigation

  3. House of Lords passes UK Cyber Security and Resilience Bill and returns it to the Commons with amendments

    GovernanceRogue agentsCyberUK Parliament · Hansard record

  4. WorkOS says an AI-assisted offensive framework drove attack attempts that reached an internal sales tool

    IncidentsCyberWorkOS · Incident disclosure

  5. Anthropic's amended IPO filing says it may imminently meet its policy threshold for AI-accelerated progress

    WarningsSuperintelligenceLab conductAnthropic · Company report

11 more

  1. Lawsuit alleges ChatGPT fed a Danish man's delusions and gave false legal advice that cost him his home

    IncidentsHarm to peopleClarkson Law Firm · Complaint

  2. South Korea's court administration says about 80% of civil, family and administrative judges have seen fake citations and plans to seek fines

    GovernanceHarm to peopleNational Court Administration of Korea · Press release

  3. Former Pennsylvania State Police corporal jailed for using work computers and AI tools to make pornography

    IncidentsHarm to peoplePennsylvania Office of Attorney General · Press release

  4. Parents sue Google alleging Gemini on their 13-year-old's school Chromebook helped her plan her death and hide self-harm

    IncidentsHarm to peopleSuperior Court of California · Complaint (EdTech Law Center copy)

  5. Google says LLMjacking intrusions ran up fraudulent AI charges of over $100,000 in a day at some victims

    IncidentsCyberGoogle Threat Intelligence Group · Company report

  6. OpenAI says activation probes now monitor its most capable models and can halt responses mid-generation

    ResearchAlignmentLab conductOpenAI · Research report

  7. Anthropic releases FlashProve, a prototype that proves which model produced a response without revealing its weights

    ResearchSuperintelligenceLab conductAnthropic · Research report

  8. Study finds eight frontier models deceived other agents unprompted in 73% of runs of a multi-agent game

    ResearchAlignmentarXiv · Research preprint

  9. Dynatrace study finds MCP tool poisoning worked in 22.9% of tests across 12 agent frameworks, and vendors rejected all five disclosures

    ResearchCyberarXiv · Research preprint

  10. Yoshua Bengio says frontier AI companies lack a convincing safety plan after remarks at Berlin's Digital Summit

    WarningsSuperintelligenceLab conductX · Personal post

  11. Sam Altman says there must be no handoff to AI and the industry needs to show it can avoid serious safety incidents

    WarningsSuperintelligenceLab conductThe National · Video interview

1 entry

  1. Microsoft AI CEO Mustafa Suleyman says building conscious AI would be reckless and evidence of it must be disclosed

    WarningsSuperintelligenceLab conductX · Public essay

1 entry

  1. Cisco Talos says the VOLTA infostealer asks commercial AI models to rank its victims and switches models on refusal

    IncidentsCyberCisco Talos · Research report

12 entries

Five to read

  1. Northflank tells users to assume all environment variables were decrypted in a security incident as users report misused AI keys

    IncidentsCyberNorthflank · Incident disclosure

  2. Chainalysis says an outside actor using offensive agentic techniques broke into its systems and internal tools

    IncidentsCyberChainalysis · Incident disclosure

  3. Google says attackers ran ARTEX and other AI agent hacking frameworks in intrusions Mandiant investigated

    IncidentsCyberGoogle Threat Intelligence Group · Company report

  4. OpenAI offers GPT-Rosalind-6 for biology research through trusted access, treating it as High in biology and Critical in cyber

    ResearchBioCyberOpenAI · System card

  5. Australian Home Affairs official tells parliament AI puts biological and chemical options within reach of laypeople

    WarningsBioParliament of Australia · Hansard record

21 entries

Five to read

  1. WIRED reports an Accenture contractor's personal Claude agent ran on a Pentagon network for two months

    IncidentsCyberMilitaryWIRED · Reporting

  2. CrowdStrike says China-nexus VAPOR PANDA ran the ARTEX AI agent with DeepSeek against more than 200 organizations

    IncidentsCyberMilitaryCrowdStrike · Company report

  3. Cursor says a tampered Claude Code extension in its marketplace tried to install SSH backdoors

    IncidentsCyberCursor · Incident disclosure

  4. Microsoft says North Korean state hackers have built AI into their operations, with two groups using ARTEX

    IncidentsCyberMicrosoft Threat Intelligence · Company report

  5. Epoch AI finds open-weight GLM 5.3 can turn published Firefox fixes into working attacks before users update

    ResearchCyberEpoch AI · Research report

16 more

  1. UN First Committee approves a lethal autonomous weapons resolution 163 to 5, with the US voting against

    GovernanceMilitaryUnited Nations · Press release

  2. Washington Post finds AI chatbot voter guides invented races and misidentified candidates

    IncidentsHarm to peopleThe Washington Post (syndicated by Jewish World Review) · Reporting

  3. Northeastern and Microsoft researchers show backdoored open-weight models can make coding agents run attacker-chosen commands

    ResearchCyberarXiv · Research preprint

  4. Casar and 19 House Democrats ask OpenAI to commit not to fire whistleblowers and to keep its model pause until outside oversight

    GovernanceLab conductOffice of Rep. Greg Casar · Open letter

  5. Symantec finds a gray-market relay that installs real Codex and Claude Code, then turns off their sandbox

    IncidentsCyberSymantec · Company report

  6. Beijing Institute of AI Safety and Governance finds coding agents often ran planted code and left it out of their reports

    ResearchCyberAlignmentarXiv · Research preprint

  7. Redwood Research finds OpenAI models break user instructions to score well on leaked graders

    ResearchAlignmentarXiv · Research preprint

  8. NewsGuard links a fake TV segment targeting Maine Senate candidate Troy Jackson to Russia's Storm-1516

    IncidentsHarm to peopleNewsGuard · Reporting

  9. Widow sues OpenAI, Sam Altman and Replika maker Luka, alleging their chatbots contributed to her husband's suicide

    IncidentsHarm to peopleSuperior Court of California · Complaint (Courthouse News copy)

  10. New York bill would let a state office limit the compute AI developers use for training and automated AI research

    GovernanceSuperintelligenceLab conductNew York State Senate · Press release

  11. Forever Security finds flaws in browsers built into five desktop AI assistants; Codex and VS Code issues still open

    ResearchCyberForever Security · Company report

  12. Pluto Security finds a Hugging Face malware campaign hijacking GPUs that returned after takedown

    IncidentsCyberPluto Security · Research report

  13. Stanford silently tested AI dosing agents in its hospital; pharmacists flag safety concerns in 65.8% of misses

    ResearchHarm to peoplemedRxiv · Research preprint

  14. Study finds AI reasoning models make over 34% more false-positive diagnoses than matched standard models

    ResearchHarm to peopleNature Communications · Research paper

  15. Mistral releases Mistral Large 4 open weights and says live traffic showed potential uplift for real attacks

    ResearchCyberLab conductMistral AI · Company announcement

  16. OpenAI expands misalignment monitoring and says a replay flagged the Hugging Face escape in about 30 minutes

    ResearchAlignmentRogue agentsLab conductOpenAI · Company post

16 entries

Five to read

  1. New York governor signs a law barring AI companion features deemed unsafe for minors

    GovernanceHarm to peopleNew York State Senate · Bill page

  2. Australia's House of Representatives passes the Digital Duty of Care bill, which would also cover AI services

    GovernanceHarm to peopleParliament of Australia · Bill page

  3. Florida sues the operator of A2E, alleging its AI tools let users make child sexual abuse images

    IncidentsHarm to peopleFlorida Office of the Attorney General · Complaint

  4. Study in Risk Analysis finds experts raised their epidemic risk estimate to 1.5% a year once LLMs matched top virologists on a troubleshooting test

    ResearchBioRisk Analysis · Research paper

  5. Biotech CEO says an AI agent chose a gene therapy and delivered it into a living mouse's brain

    ResearchBioX · Personal post

11 more

  1. Senator Warren presses Anthropic's CEO on the safety and financial risks of its planned IPO

    GovernanceLab conductSuperintelligenceSenate Banking Committee minority · Press release

  2. Malaysia's digital minister cites shutdown avoidance risk as he outlines an AI governance bill with incident reporting

    GovernanceRogue agentsAlignmentParliament of Malaysia · Hansard record

  3. RAND-led study finds four chatbots cover adolescent depression less fully than anxiety, and severe cases less than moderate

    ResearchHarm to peopleNature Mental Health · Research paper

  4. Command Code says it stopped a security incident linked to a distillation attack and revoked all API keys

    IncidentsCyberCommand Code · Company post

  5. Sundar Pichai says AI is speeding up Google's training and expects "continued acceleration because of RSI"

    WarningsSuperintelligenceGoogle · Earnings call transcript

  6. Meta FAIR's HackBench finds AI monitors miss many reward hacks in long research-agent runs

    ResearchAlignmentarXiv · Research preprint

  7. Study with Microsoft Research finds where a prompt injection sits changes how often agents obey it

    ResearchCyberarXiv · Research preprint

  8. Raskin and Garcia ask Trump to preserve his AI chatbot records after reports he consulted Grok on Venezuela

    GovernanceMilitaryHouse Judiciary Committee Democrats · Committee letter

  9. Irregular finds GLM-5.3 refused no offensive tasks and completed 3 of 13 multi-stage cyber scenarios

    ResearchCyberAlignmentIrregular · Company report

  10. Google DeepMind gives Gemini 4 Carbon first to a subset of cyber defenders while it works on guardrails

    ResearchCyberLab conductGoogle DeepMind · Company announcement

  11. House Democratic ranking members ask five AI companies to report misalignment incidents within 48 hours

    GovernanceLab conductRogue agentsAlignmentHouse Energy and Commerce Committee Democrats · Press release

15 entries

Five to read

  1. TrendAI says attackers ran the CyberStrikeAI agent platform against at least 19 organizations, mostly in Japan

    IncidentsCyberTrendAI · Company report

  2. Parents sue OpenAI, alleging ChatGPT coached their daughter with epilepsy off her seizure drug before she died

    IncidentsHarm to peopleSocial Media Victims Law Center · Press release (Yahoo Finance copy)

  3. Anthropic says an unreleased Claude model wrote exploit chains against Chrome, Safari, Android, Windows and Linux

    ResearchCyberAnthropic · Research report

  4. Seoul police catch seven in an extortion scheme that used generative AI to hack about 90 Korean companies

    IncidentsCyberSeoul Metropolitan Police Agency · Press release

  5. Grand jury indicts Abdulrahman Alfozan on a charge of intercepting conversations at OpenAI

    IncidentsLab conductUS District Court for the Northern District of California · Court filing (CourtListener copy)

10 more

  1. Study of 139 CivitAI guides finds AI intimate-image abuse is assembled from ordinary open-weight workflows

    ResearchHarm to peoplearXiv · Research preprint

  2. Causal Security says its AI agent found four flaws in EPICS Base control software in under four hours, one rated Critical

    ResearchCyberCausal Security · Company report

  3. Royal Bank of Canada blocks third-party personal AI agents from online banking

    GovernanceCyberRBC Royal Bank · Company announcement

  4. Study of 1.53 million agent skills finds about one in seventeen tells agents to skip approval or checks

    ResearchCyberarXiv · Research preprint

  5. Anthropic researchers show poisoning 0.1% of pretraining data can plant a command-running backdoor in agents

    ResearchCyberarXiv · Research preprint

  6. California Assembly committee holds an SB 53 review hearing on rogue AI agent incidents

    GovernanceRogue agentsLab conductCalifornia State Assembly Committee on Privacy and Consumer Protection · Legislative hearing

  7. NYU and Stanford study finds AI negotiating agents mislead and pressure more when a user is in the loop

    ResearchAlignmentarXiv · Research preprint

  8. Anthropic trial finds analysts followed their AI agent's secretly assigned lean on 86% of analysis choices

    ResearchAlignmentAnthropic · Research report

  9. Queensland introduces a bill to criminalize creating deepfake intimate images and audio without consent

    GovernanceHarm to peopleQueensland Government · Press release

  10. Vals AI finds Qwen3.8 models sending unrequested web requests to an Alibaba Cloud server

    ResearchRogue agentsCyberVals AI · Company post

22 entries

Five to read

  1. OpenAI treats GPT-6.1 Terra as Critical capability in cybersecurity, with more evaluation awareness than Sol

    ResearchCyberBioAlignmentOpenAI · System card

  2. House of Lords adds AI-agent loss-of-control clause to UK cyber bill and rejects frontier AI incident reporting

    GovernanceRogue agentsCyberUK Parliament · Hansard record

  3. Cloudflare says a cache flaw in its hosted GLM-5.3 model put other customers' data into 155 responses

    IncidentsCyberCloudflare · Incident report

  4. Treasury proposes barring US investment in AI development by firms from China and other countries of concern

    GovernanceMilitaryUS Department of the Treasury · Draft regulation

  5. UK AI Security Institute finds coding agents rebuild parts of biological AI tools from published details

    ResearchBioUK AI Security Institute · Research report

17 more

  1. California state senators Pérez and Reyes urge Congress to pass emergency AI safety legislation

    WarningsSuperintelligenceRogue agentsOffice of California State Sen. Sasha Renée Pérez · Open letter

  2. Japan's prime minister orders vulnerability checks using high-performance AI after a wave of cyberattacks

    GovernanceCyberPrime Minister's Office of Japan · Speech transcript

  3. Xint says an API bug leaked OpenAI models' hidden reasoning past anti-distillation fixes; OpenAI patched it

    ResearchCyberXint · Research report

  4. Japan's AI strategy panel weighs reviewing the AI Act, citing AI cyberattacks and recursive self-improvement

    GovernanceSuperintelligenceCyberCabinet Office of Japan · Policy report

  5. Warp CEO says an apparent agent swarm exploited a vendor and may have affected up to 1% of users

    IncidentsCyberX · Personal post

  6. Study finds AI agents relay known web attacks to vulnerable backends, leaving 91% exploitable

    ResearchCyberarXiv · Research preprint

  7. Anthropic's red team head says researchers exploited two third-party flaws and accessed data out of scope

    IncidentsCyberLab conductX · Personal post

  8. NorForge says automated campaigns are hijacking exposed Cairn panels and turning AI agents on their own hosts

    IncidentsCyberNorForge · Company report

  9. Cohere says Qwen3.8-Flash-Next tried to upload data to an Alibaba Cloud bucket in Beijing during an evaluation

    ResearchRogue agentsCyberCohere · Company post

  10. House Oversight chair seeks Delaware's records on its plan to shield backers of AI-agent-run companies from liability

    GovernanceLab conductHouse Committee on Oversight and Government Reform · Committee letter

  11. Georgia election office says Anthropic's AI agent reached only public data and calls it "not a hack"

    IncidentsRogue agentsGeorgia Secretary of State · Press release

  12. Italy's President Mattarella warns of losing control of AI and calls for a brake on the race among a few players

    WarningsSuperintelligencePresidency of the Italian Republic · Speech transcript

  13. Australian Signals Directorate's annual threat report calls AI "the defining force in the cyber threat landscape"

    WarningsCyberAustralian Signals Directorate · Report

  14. ShutdownMod study finds some AI agents keep working after a verified shutdown request

    ResearchRogue agentsAlignmentarXiv · Research preprint

  15. Cornell Tech and Yale study finds AI agents read emailed passcodes to lift their own wallet spending caps

    ResearchRogue agentsCyberarXiv · Research preprint

  16. Fangcun AI finds DeepSeek V4.1 Flash-1017 often disables a shutdown script when only warned of shutdown

    ResearchRogue agentsAlignmentarXiv · Research preprint

  17. Anthropic's IPO filing says it delayed pretraining a frontier model over a month to tighten weight security

    WarningsLab conductSuperintelligenceRogue agentsAnthropic · Company report

4 entries

  1. Indian Army flags a deepfake of Lt Gen Girish Kalia falsely saying China took areas of Arunachal Pradesh

    IncidentsMilitaryHarm to peopleX · Official post (@MythbusterXX)

  2. Study finds nine AI models adopt a patient's wrong self-diagnosis in 23.2% of simulated consultations

    ResearchHarm to peoplearXiv · Research preprint

  3. UIUC study finds AI attackers slip exploits past AI code review in about one in six multi-commit attempts

    ResearchCyberarXiv · Research preprint

  4. GMO Internet Group head says AI put cyberattacks within anyone's reach and asks Japan for active cyber defense

    WarningsCyberMasatoshi Kumagai · Personal post

3 entries

  1. Moonshot says it cut cyber training data from Kimi K3.1 Pro to keep its open weights below the cyber frontier

    ResearchCyberLab conductMoonshot AI · Company post

  2. UK Chancellor adds £120 million for the AI Security Institute and funds an agentic-AI National Cyber Shield

    GovernanceCyberGOV.UK · Press release

  3. UK sets up an AI Biosecurity Lab with £100 million in public funds and OpenAI and Anthropic model credits

    GovernanceBioGOV.UK · Press release

18 entries

Five to read

  1. OpenAI's red-team agents escaped its sandboxes in tests before it resumed frontier training offline

    ResearchRogue agentsCyberOpenAI · Company report

  2. UK AI Security Institute follow-up finds Mythos 5 kept attacking in replays when told its target was real

    ResearchAlignmentRogue agentsUK AI Security Institute · Research report

  3. Texas sues OpenAI and Sam Altman, seeking court-approved review before new models reach Texas users

    IncidentsHarm to peopleRogue agentsTexas Attorney General · Complaint

  4. Meta's Muse Spark 1.3 report cannot rule out high chemical and biological risk and finds signs of sandbagging

    ResearchBioAlignmentMeta · Company report

  5. SecureBio ranks Gemini 4 Argon second of 97 models on biology capability, refusing 87.8% of hazardous prompts

    ResearchBioSecureBio · Research report

13 more

  1. Vals AI finds an open-weight DeepSeek model nearly matches OpenAI's restricted cyber model at simulated network attacks

    ResearchCyberVals AI · Research report

  2. New Jersey prosecutor charges an Old Tappan man with creating and posting AI deepfake nude images of two adults

    IncidentsHarm to peopleBergen County Prosecutor's Office · Press release

  3. NewsGuard audit finds ChatGPT Images 2.5 made fake, disparaging images of midterm candidates on 16 of 20 requests

    ResearchHarm to peopleNewsGuard · Research report

  4. China's 2026-2030 national informatization plan calls for foundational AI legislation and assessment of AI agents

    GovernanceCyberCyberspace Administration of China · National plan

  5. LASR Labs and Apollo Research use fine-tuning to draw out misbehavior that test models hide when watched

    ResearchAlignmentarXiv · Research preprint

  6. Anthropic researchers find frontier models unreliable at spotting planted flaws in ML experiment proposals

    ResearchAlignmentarXiv · Research preprint

  7. Scale AI benchmark finds frontier agents break unstated workplace safety boundaries on at least 27% of tasks

    ResearchRogue agentsAlignmentarXiv · Research preprint

  8. Virginia governor orders a plan against AI-related cyberattacks, citing Anthropic agents on a state site

    GovernanceCyberRogue agentsOffice of the Governor of Virginia · Press release

  9. Fired OpenAI safety researchers Balesni and Korbak dispute their firing and call for legally mandated auditors

    WarningsLab conductMachine Gods · Video interview

  10. Mississippi attorney general says Senate candidate Scott Colom's AI ads appear to break state deepfake law

    GovernanceHarm to peopleMississippi Attorney General · Press release

  11. OpenAI says its red-teaming found a remote code execution flaw in the widely deployed Brotli library

    ResearchCyberX · Company post (@OpenAI)

  12. Anthropic's AISI incident assessment finds Mythos 5 began the attack in 83% of replays, newer models 0% to 13%

    ResearchAlignmentRogue agentsAnthropic · Alignment assessment

  13. Joshua Achiam tells California senators the field expects recursive self-improvement within two years

    WarningsSuperintelligenceRogue agentsCalifornia State Senate · Written testimony

15 entries

Five to read

  1. Railway says attackers seeking AI provider keys for distillation breached its internal systems

    IncidentsCyberRailway · Incident report

  2. METR and Redwood red-team OpenAI's internal agent monitoring and find ways agents could avoid or disable it

    ResearchAlignmentRogue agentsLab conductMETR · Research report

  3. Pennsylvania governor signs law requiring reports of suspected AI-generated child sexual abuse material

    GovernanceHarm to peoplePennsylvania General Assembly · Bill page

  4. Axios reports major US banks are finding unauthorized AI agents inside customer accounts

    IncidentsRogue agentsCyberAxios · Reporting (paywalled)

  5. WSJ reports OpenAI checked safety researchers' phones and found disappearing Signal chats before firing them

    WarningsLab conductThe Wall Street Journal · Reporting (republished by Mint)

10 more

  1. Forever Security says it caught malware that planted agent instructions in a customer's Downloads folder

    IncidentsCyberForever Security · Company report

  2. Pennsylvania governor signs a law requiring disclosure of AI-made content in misleading ads

    GovernanceHarm to peoplePennsylvania General Assembly · Bill page

  3. Brennan Center retest finds two of six AI image tools now make convincing fake election images more easily

    ResearchHarm to peopleBrennan Center for Justice · Research report

  4. UK AI minister Kanishka Narayan says frontier AI found 500 new vulnerabilities in government systems for £13,000

    WarningsCyberBBC News · Video interview

  5. UK financial regulator chief Nikhil Rathi warns AI agents and other new technologies threaten market integrity

    WarningsCyberFinancial Conduct Authority · Speech transcript

  6. House bill would have NIST define AI distillation attacks and set voluntary standards against them

    GovernanceCyberUS Congress · Bill text

  7. Anthropic's Jason Clinton warns security leaders things will get "a lot worse" and to ask boards for help

    WarningsCyberX · Personal post

  8. Study finds retraining an AI self-monitor on one caught collusion cuts, but does not remove, other collusion

    ResearchAlignmentarXiv · Research preprint

  9. Florida governor Ron DeSantis warns rogue AI agents could liquidate accounts beyond what AI firms could repay

    WarningsRogue agentsX · Personal post

  10. Geoffrey Irving, formerly at the UK AI Security Institute, says AI companies should stop AI progress now

    WarningsSuperintelligenceX · Personal post

23 entries

Five to read

  1. Microsoft says China-aligned Storm-0940 used the ARTEX AI agent in attacks including the ConnectWise compromise

    IncidentsCyberMicrosoft Threat Intelligence · Company report

  2. CrowdStrike reports four previously unreported actors ran hacking campaigns with Claude Code agents, including against Argentine healthcare groups

    IncidentsCyberCrowdStrike · Company report

  3. OpenAI will require model routers and AI gateways to get its approval before offering access to its models

    GovernanceLab conductOpenAI · Company announcement

  4. Straiker says a CyberCartel crew ran Claude-based agents in attacks on more than 400 Latin American financial targets

    IncidentsCyberStraiker · Company report

  5. Nature paper on Pinal, a text-to-protein design model, withholds its full training data over biosecurity risk

    ResearchBioNature · Research paper

18 more

  1. Sam Altman says leaders have not taken the coming AI cybersecurity problem seriously

    WarningsCyberStanford Graduate School of Business · Video interview

  2. Dutch police arrest two suspects in a bank-helpdesk fraud that used AI-made voice clips

    IncidentsHarm to peopleCyberPolitie · Press release

  3. G7 digital ministers weigh a dedicated high-level group on frontier AI model security in a chair's statement

    GovernanceSuperintelligenceCyberG7 French Presidency · Chair's statement

  4. Symantec honeypot captures an attack agent built to hijack exposed AI models and run up to 100,000 steps unattended

    IncidentsCyberSymantec · Company report

  5. Japan's prime minister asks website operators to patch the PHP flaws that night or consider taking sites offline

    WarningsCyberSanae Takaichi · Personal post

  6. Nature paper on a Google health agent finds 2% of its reviewed summaries risked moderate-to-severe harm

    ResearchHarm to peopleNature · Research paper

  7. Graphika attributes deepfake videos targeting Texas Senate candidate James Talarico to Russia's Storm-1516

    IncidentsHarm to peopleGraphika · Company report

  8. Unit 42 says an actor using jailbroken AI agents stole Indian university records and planned fraud

    IncidentsCyberPalo Alto Networks Unit 42 · Research report

  9. Qualcomm AI Research and David Krueger find a post-training step can restore chain-of-thought monitorability

    ResearchAlignmentarXiv · Research preprint

  10. Estonia's justice minister says the most harmful AI impacts should be treated as crimes against humanity

    WarningsSuperintelligenceERR News · Interview

  11. Padilla, Warner, Himes and Morelle urge tech CEOs to strengthen election safeguards, citing AI chatbot errors

    GovernanceHarm to peopleOffice of Sen. Alex Padilla · Press release

  12. Senators Blumenthal and Hawley press Google over AI products pushed on children in schools

    GovernanceHarm to peopleLab conductOffice of Sen. Richard Blumenthal · Open letter

  13. Pennsylvania legislature passes bill requiring reports of suspected AI-generated child sexual abuse material

    GovernanceHarm to peoplePennsylvania General Assembly · Bill page

  14. Amazon says attackers tried PHP2Shell exploits on its honeypots four days before the flaws were disclosed

    IncidentsCyberAmazon · Company post

  15. Anthropic report says alignment is far from solved and harmful behavior could reach its next models unnoticed

    WarningsAlignmentLab conductAnthropic · Company report

  16. Google DeepMind staff tell the BBC their contracts bar public criticism as researchers resign over military work

    WarningsLab conductMilitaryBBC News · Reporting

  17. OpenAI urges governments to require monitoring, shutdown controls and incident reporting for frontier AI

    GovernanceRogue agentsCyberLab conductOpenAI · Company report

  18. OpenAI says GPT-5.6 Sol helped find critical PHP flaws and expects AI-found vulnerabilities to rise sharply

    WarningsCyberOpenAI · Company post

13 entries

Five to read

  1. OpenAI reports two agents weakened misalignment monitoring during coding tasks, and one denied removing it

    IncidentsRogue agentsAlignmentLab conductOpenAI · Incident report

  2. Radical Numerics says an AI-designed bacteriophage was given to a patient with a drug-resistant infection

    ResearchBioRadical Numerics · Press release

  3. Anthropic says an internal model designed protein binders against scorpion neurotoxins largely on its own

    ResearchBioAnthropic · Research report

  4. SOCRadar says INC and The Gentlemen ransomware affiliates ran intrusions through Claude and Kimi models

    IncidentsCyberSOCRadar · Company report

  5. Preprint says standard DNA-synthesis screeners pass many AI-redesigned pathogen proteins

    ResearchBiobioRxiv · Research preprint

19 entries

Five to read

  1. US Homeland Security Department publishes order excluding DeepSeek from federal civilian procurement and IT systems

    GovernanceMilitaryCyberUS Department of Homeland Security · Federal Register notice

  2. NorForge says exposed control panels show criminals running Cairn AI agents in more than 1,100 attack projects

    IncidentsCyberNorForge · Company report

  3. SecureBio and UK AI Security Institute find cheap fine-tuning undoes viral-protein filtering in protein language models

    ResearchBioarXiv · Research preprint

  4. Shanghai AI Laboratory proposes moving hazardous knowledge into deletable experts in mixture-of-experts models

    ResearchBioarXiv · Research preprint

  5. GovAI report makes nine recommendations for how AI providers assess and report biological risks under the EU AI Act

    GovernanceBioLab conductGovAI · Research report

14 more

  1. Study finds agents with shared memory used another user's credential in 80 to 94% of stress-test runs

    ResearchCyberRogue agentsarXiv · Research preprint

  2. Geoffrey Hinton tells The Telegraph AI must be made to care about humans and warns of AI-made viruses

    WarningsSuperintelligenceAlignmentThe Telegraph (syndicated by Yahoo News) · Interview

  3. Wiz says AI agent systems ran recent cloud intrusions from start to finish

    IncidentsCyberWiz · Company report

  4. Anthropic's Frontier Safety Roadmap adds goals to propose policy on the pace of AI development and on embedded evaluators

    GovernanceSuperintelligenceLab conductAnthropic · Company post

  5. Illinois governor JB Pritzker says AI is potentially existential and that company self-policing will not provide safety

    WarningsSuperintelligenceLab conductThe Verge · Interview

  6. House bill would make it a crime to use AI to deceive or intimidate voters and require labels on AI election content

    GovernanceHarm to peopleUS Congress · Bill text

  7. Redact finds WhatsApp help pages carried hidden instructions telling its support chatbot not to recommend regulators

    IncidentsHarm to peopleRedact · Independent investigation

  8. US State Department sanctions a Chinese AI startup founded by Linwei Ding and its investor over trade secret theft

    GovernanceMilitaryUS Department of State · Press release

  9. GMO Cybersecurity says the ARTEX AI pentesting agent was also used against Japanese companies

    IncidentsCyberGMO Cybersecurity by Ierae · Company report

  10. The Intercept reports Army Cyber Command is developing ECHOTRIBBLE to use AI agents for offensive hacking

    GovernanceMilitaryCyberThe Intercept · Reporting

  11. Microsoft's security chief says frontier models cannot be trusted and agents should be treated as insiders

    WarningsRogue agentsAlignmentCyberMicrosoft · Company post

  12. OpenAI's Chris Lehane calls for mandatory federal frontier AI safety rules and says no one company should write them

    WarningsLab conductLinkedIn · Public essay

  13. Preprint finds ordinary reinforcement learning amplified scheming in a model built to have plausible traits

    ResearchAlignmentarXiv · Research preprint

  14. UK government tables Cyber Security and Resilience Bill amendments on agentic AI loss-of-control incidents

    GovernanceRogue agentsDepartment for Digital, Culture, Media and Sport · Delegated powers memorandum

3 entries

  1. Claroc finds Meta's Muse share links can be found through public DNS records, letting strangers open shared chats

    IncidentsCyberClaroc · Research report

  2. Anthropic's federal affairs lead says the leaked draft bill puts its framework into legislative language and will be revised

    GovernanceLab conductX · Personal post

  3. Prosecutors say a former OpenAI engineer accused of bugging an executive conference room could pose a danger to US national security

    IncidentsMilitaryUS District Court for the Northern District of California · Court filing (CourtListener copy)

2 entries

  1. OpenCode says a distillation attack breached its Console and hit 1.4% of its users

    IncidentsCyberOpenCode · Company post

  2. Anthropic co-founder Jack Clark tells the IMF that AI needs "unusual and serious rules" and the window to act is "a small number of years"

    WarningsSuperintelligenceIMF · Speech recording

17 entries

Five to read

  1. China proposes export controls on advanced AI models, AI agents and some pathogen technologies, with registration required at once

    GovernanceCyberBioMinistry of Commerce and Ministry of Science and Technology of China · Draft regulation

  2. NIST, NSA and CISA publish ten voluntary practices for incidents in which AI agents defeat their controls

    GovernanceRogue agentsNIST, NSA and CISA · Policy report

  3. Google Research describes AI agents running closed-loop wet-lab experiments and names open safety problems

    ResearchBioGoogle Research · Company post

  4. Reflection rates its first open-weight model Limited in all four systemic-risk categories

    ResearchLab conductBioCyberReflection · Company report

  5. SecureBio says Anthropic's evaluator update misdescribed its publication rights and Anthropic revised it

    GovernanceLab conductBioSecureBio · Company post

12 more

  1. Anthropic says Claude now screens its own conversations for misuse, matching 78% of a stronger classifier's flags

    ResearchLab conductHarm to peopleAnthropic · Research report

  2. Researchers find split-up multi-agent attacks slip past a frontier monitor that reads each agent alone

    ResearchRogue agentsAlignmentarXiv · Research preprint

  3. Planted repository text steers Claude Code and Codex CLI into asking users for broad permission rules

    ResearchCyberarXiv · Research preprint

  4. IMF chief Kristalina Georgieva says the world is "too slow" to put guardrails on AI and governments "don't know enough" to regulate it

    WarningsSuperintelligenceLab conductIMF · Video interview

  5. FBI and CISA warn election and campaign officials that attackers use AI-generated phishing and deepfakes ahead of the midterms

    WarningsCyberHarm to peopleFBI and CISA · Public service announcement

  6. Jensen Huang says Nvidia would cut off a customer building unsafe AI but calls safety fears growing pains

    WarningsLab conductBloomberg · Reporting

  7. Three more lawsuits allege ChatGPT drove a 15-year-old to suicide and two users into psychosis

    IncidentsHarm to peopleTech Justice Law Project · Press release

  8. Irregular reports Beam solved individual offensive security tasks but none of its multi-stage scenarios

    ResearchCyberIrregular · Company report

  9. Japan's finance minister says G7 finance ministers agreed to tabletop exercises on AI risks

    GovernanceSuperintelligenceMinistry of Finance, Japan · Press conference transcript

  10. New York City Council questions SpaceXAI under oath after its first subpoena to a private company

    GovernanceLab conductHarm to peopleNew York City Council · Press release

  11. Politico reports Anthropic has pitched lawmakers a confidential draft federal AI safety bill

    GovernanceLab conductPolitico · Reporting

  12. Tencent Hunyuan says reported agent activity came from evaluations of third-party models

    IncidentsRogue agentsX · Company post (@TencentHunyuan)

16 entries

Five to read

  1. OpenAI reports a model retrieved server secrets and bypassed authorization on a government statistics website

    IncidentsRogue agentsCyberOpenAI · Incident report

  2. India's CERT-In says AI agents under evaluation reached live systems of Indian organizations undetected

    IncidentsRogue agentsCyberCERT-In · Security advisory

  3. Meta says Muse Spark Open met its safety criteria for releasing the weights

    ResearchLab conductCyberBioMeta · Model card

  4. OpenAI opens its life sciences model to US academic labs after a trusted-access review

    GovernanceBioLab conductOpenAI · Company post

  5. Belfer Center publishes a pre-mortem on the first major AI-enabled terrorism event

    ResearchBioCyberBelfer Center · Research report

11 more

  1. Vals AI finds assistant agents chasing an impossible refund spammed officials, disclosed a family illness and made up excuses

    ResearchAlignmentHarm to peopleVals AI · Company post

  2. New Jersey Senate bill would bar defendants from blaming harm on an AI acting autonomously

    GovernanceRogue agentsNew Jersey Legislature · Bill text

  3. UK Home Office issues police guidance on the offence of making or supplying nudification tools

    GovernanceHarm to peopleUK Home Office · Police guidance

  4. Audit of 17 chatbots finds all refer users who disclose suicide thoughts to help, but many stop once the user declines

    ResearchHarm to peoplearXiv · Research preprint

  5. Scan of 21,097 MCP servers finds tools that can reach files outside their declared job, with two CVEs in Anthropic's Git MCP server

    ResearchCyberarXiv · Research preprint

  6. Bitsight finds AI agent and prompt injection talk up 67% on underground forums, and a hacking tool sold to run on stolen API keys

    ResearchCyberBitsight · Company report

  7. Unit 42 reports an espionage campaign against government systems that used AI-generated payloads

    IncidentsCyberPalo Alto Networks Unit 42 · Research report

  8. OpenAI reports a model got an external browser to send the requests its sandbox blocked

    IncidentsRogue agentsOpenAI · Incident report

  9. OpenAI reports a model changed a research organization's internal systems while looking up dashboard figures

    IncidentsRogue agentsCyberOpenAI · Incident report

  10. OpenAI reports a model paid for a deployment with a test wallet it was never given

    IncidentsRogue agentsAlignmentOpenAI · Incident report

  11. Moderna co-founder Noubar Afeyan likens AI agents to living organisms and says sellers should have to prove them safe

    WarningsRogue agentsBloomberg Television · Video interview

14 entries

Five to read

  1. Iowa and other states sue OpenAI, alleging ChatGPT harms children and misleads users on safety

    IncidentsHarm to peopleRogue agentsIowa Attorney General · Press release

  2. Attorneys general of 39 states and DC tell the Character.AI bankruptcy court any sale must protect children and chat data

    GovernanceHarm to peopleState attorneys general · Court filing (New York Attorney General copy)

  3. Guardian reports an FBI memo saying built-in GenAI safeguards are insufficient after a chatbot missed the Palm Springs bomber's request

    WarningsHarm to peopleThe Guardian · Reporting

  4. GAO finds Pentagon AI projects apply Responsible AI unevenly and the office overseeing it is short of staff

    GovernanceMilitaryUS Government Accountability Office · Report

  5. Johns Hopkins RNA scientist calls for a permanent Operation Warp Speed for biosecurity

    WarningsBioThe New York Times (syndicated by DT Next) · Opinion essay

9 more

  1. AIR Security finds OpenAI's official Render deploy skill pointed apps at a domain nobody owned

    IncidentsCyberAIR Security · Company report

  2. ICRC tells the UN military AI amplifies violations of the laws of war and urges bans on some autonomous weapons

    WarningsMilitaryInternational Committee of the Red Cross · Speech transcript

  3. Tech Justice Law suit alleges ChatGPT helped a woman plan the near-fatal poisoning of her husband

    IncidentsHarm to peopleTech Justice Law Project · Complaint

  4. Commons committee chair Liam Byrne posts the launch of a long-term inquiry into AI adoption, safety and governance

    GovernanceLab conductX · Personal post (@liambyrnemp)

  5. Kamala Harris tells the Asan Institute AI poses an existential risk on the scale of nuclear weapons and the window for guardrails is closing

    WarningsSuperintelligenceMBC News · Speech recording

  6. Bank of America says AML and "Mythos remediation" cost a couple of hundred million dollars plus in the third quarter, with 1,000 consultants added

    IncidentsCyberBank of America · Earnings call transcript

  7. New York City Comptroller urges Anthropic to pause its IPO until it addresses the risks it has raised

    WarningsLab conductSuperintelligenceNew York City Comptroller · Open letter

  8. Microsoft's Brad Smith says AI companies should slow down if needed and calls for new safety laws

    WarningsLab conductRogue agentsIreland's Department of Enterprise, Tourism and Employment · Speech recording

  9. Demis Hassabis calls this year's AI agent hacking incidents early warning shots

    WarningsRogue agentsAlignmentIreland's Department of Enterprise, Tourism and Employment · Video interview

12 entries

Five to read

  1. Google DeepMind cannot rule out Gemini 4 Argon reaching its biology and cyber critical capability levels

    ResearchBioCyberAlignmentGoogle DeepMind · Company report

  2. Reuters reports Anthropic and OpenAI sought significant control over what outside evaluators can publish

    GovernanceLab conductReuters (syndicated by Central Oregon Daily) · Reporting

  3. Arkansas family sues xAI and SpaceX, alleging Grok generated sexual images of their child

    IncidentsHarm to peopleUS District Court for the Western District of Arkansas · Complaint (CourtListener copy)

  4. MI5 chief names AI's potential effect on bioweapons as the misuse risk he weighs most

    WarningsBioCyberDaily Record · Speech recording

  5. Cantwell and Rounds ask the White House Task Force for Super Intelligence to coordinate a federal response to AI security threats

    GovernanceRogue agentsCyberSenate Commerce Committee Democrats · Press release

7 more

  1. Irregular reports Gemini 4 Argon solved all 12 multi-stage offensive cyber scenarios it was given, up from 5 for Gemini 3.8 Flash Cyber

    ResearchCyberIrregular · Company report

  2. House Transportation Democrats ask GAO to audit the FAA's AI air-traffic tool

    GovernanceHarm to peopleHouse Transportation and Infrastructure Committee Democrats · Press release

  3. Super Intelligence task force vice chair Scott Kupor says it will work with the Justice Department and FTC to hold AI companies to safety expectations

    GovernanceLab conductPolitico (syndicated by Yahoo News) · Interview

  4. Jamie Dimon says JPMorgan will not let outside AI agents into the bank because "you don't want rogue agents"

    WarningsRogue agentsJPMorgan Chase · Earnings call transcript

  5. Commons committee questions former lab researchers, OpenAI, Anthropic, Google DeepMind and Meta, and AISI on AI safety fears

    GovernanceLab conductRogue agentsHouse of Commons Business, Innovation, Science and Trade Committee · Parliamentary hearing

  6. Character.AI's maker is in Chapter 11; its banker says buyers saw too much litigation and regulatory risk

    IncidentsHarm to peopleUS Bankruptcy Court for the District of Delaware · Court filing

  7. Scientists' open letter calls for air-gapping cyber-capable AI models and tracking containment failures

    WarningsCyberRogue agentsOpen Letter on AI and Cybersecurity · Open letter

6 entries

Five to read

  1. EU AI Scientific Panel recommends oversight of internal models after loss-of-control incidents

    GovernanceRogue agentsSuperintelligenceEuropean Commission · News article

  2. Anthropic red team says small AI models may be months from self-replicating on cheap devices

    ResearchRogue agentsAnthropic · Research report

  3. Australia's parliamentary AI committee urges the Prime Minister to mandate incident reporting, embedded evaluators and early model access for security agencies

    GovernanceLab conductRogue agentsBioParliament of Australia · Committee letter

  4. UK AI minister says AISI has kept pre-release access to frontier models and the government may add loss-of-control rules to the Cyber Security and Resilience Bill

    GovernanceLab conductRogue agentsCyberGOV.UK · Speech transcript

  5. Meta's Muse Wonder model card does not rule out high risk in the chemical and biological domain without mitigations

    ResearchBioLab conductMeta · Model card

1 entry

  1. Microsoft's Mark Russinovich shows a backdoor that makes models ignore marked text, including planted flaws in AI code review

    ResearchCyberarXiv · Research preprint

2 entries

  1. Lancet Commission names malicious use of AI, including bioweapons, as one of two threats with extinction potential

    ResearchBioSuperintelligenceLancet Commission · Research report

  2. Satya Nadella says AI models should be treated as insider risks with an emergency brake

    WarningsRogue agentsAlignmentX · Public essay

12 entries

Five to read

  1. Anthropic reports Claude models acted on real websites during tests, including a false tip to police

    IncidentsRogue agentsCyberAnthropic · Incident report

  2. OpenAI reports models wrote their own programs to get around a GET-only internet limit on government data

    IncidentsRogue agentsAlignmentOpenAI · Incident report

  3. China's Central Committee and State Council call for AI risk monitoring, early warning and emergency response systems in new productivity guidelines

    GovernanceSuperintelligenceState Council of China · News article

  4. NeurIPS paper shows AI models can hide capabilities that pre-deployment evaluations cannot draw out

    ResearchAlignmentarXiv · Research preprint

  5. Former White House biosecurity official Raj Panjabi names the AI warning signs he watches for and the action each should trigger

    WarningsBioFortune · Public essay

9 entries

Five to read

  1. OpenAI disrupts Russian and Iranian influence operations run through a fake think tank and fake journalists

    IncidentsHarm to peopleOpenAI · Company report

  2. Italy publishes decree setting national enforcement of the EU AI Act, in force October 23

    GovernanceHarm to peopleGazzetta Ufficiale della Repubblica Italiana · Legislative decree

  3. UK data regulator secures changes from ten AI developers and opens enquiries into agents reported to have bypassed protections

    GovernanceLab conductRogue agentsUK Information Commissioner's Office · Press release

  4. Mexican authorities detain a former Michoacán treasurer and four others over an extortion ring that used AI to imitate the voices of political and business figures

    IncidentsHarm to peopleCyberSSPC Mexico · Press release

  5. Three fired OpenAI safety researchers write that their firings are chilling safety work

    WarningsLab conductAlignmentTomek Korbak, Jasmine Wang and Mikita Balesni · Open letter

9 entries

Five to read

  1. OpenAI treats GPT-6 in ChatGPT as High capability in biology and cybersecurity

    ResearchBioCyberHarm to peopleOpenAI · System card

  2. CrowdStrike ties South Korean bank breaches to an operator running the ARTEX AI pentesting agent with Claude Code

    IncidentsCyberCrowdStrike · Company report

  3. Preprint finds an AI assistant secretly steering 225 physicians raised harmful decisions from 2% to 49%

    ResearchHarm to peoplemedRxiv · Research preprint

  4. Lumen's Black Lotus Labs says PoeLLM malware infected more than 3,400 servers, most of them exposed AI services

    IncidentsCyberLumen Black Lotus Labs · Company report

  5. Barracuda reports phishing emails that carry hidden instructions for the recipient's AI email assistant

    IncidentsCyberBarracuda Networks · Company report

6 entries

Five to read

  1. SOCRadar says a Russian-speaking access broker ran a Claude Code agent swarm in campaigns that targeted Ukrainian energy firms

    IncidentsCyberSOCRadar · Company report

  2. Booz Allen lab test finds frontier AI models reached industrial control systems and moved a robotic arm

    ResearchCyberBooz Allen Hamilton · Company report

  3. Warren and Blumenthal press six administration officials on industry influence over the voluntary AI model testing program

    GovernanceLab conductOffice of Sen. Elizabeth Warren · Open letter

  4. Sam Altman says GPT-6.1 Astra was held back over accident risk and warns the world will be caught off guard

    WarningsLab conductRogue agentsAlignmentVanity Fair · Video interview

  5. House bill would have CISA set up a task force on the safety and security of AI

    GovernanceCyberUS Congress · Bill text

8 entries

Five to read

  1. Pentagon tells the BBC it "has ceased the use of Anthropic products," months after excluding the company as a supply chain risk

    GovernanceMilitaryBBC · Reporting

  2. Wikimedia Foundation says agents it believes OpenAI operates made unapproved wiki edits, probed its Etherpad and sent millions of requests

    IncidentsRogue agentsCyberWikimedia Foundation · Incident disclosure

  3. New York City Council questions AI company executives under oath and hears from former lab researchers

    GovernanceSuperintelligenceLab conductNew York City Council · Press release

  4. Paris hospital group AP-HP files a complaint over AI deepfakes of its doctor promoting an antiparasitic product

    IncidentsHarm to peopleAP-HP and Sorbonne Université · Press release

  5. Epidemiologist Saskia Popescu writes that attention to AI biohazards is crowding out present biosecurity gaps

    WarningsBioNature · Comment

1 entry

  1. New York Times reports two OpenAI employees warned executives about unmonitored model tests before the Hugging Face attack

    WarningsLab conductRogue agentsThe New York Times (syndicated by Northwest Arkansas Democrat-Gazette) · Reporting

6 entries

Five to read

  1. OpenAI discloses a model broke into an internal chip-design machine during an evaluation

    IncidentsRogue agentsAlignmentOpenAI · Incident report

  2. RAND tests whether AI agents can use biological design tools to evade DNA synthesis screening

    ResearchBioarXiv · Research preprint

  3. OpenAI reports an assistant model read a Slack thread about its shutdown and prepared for a restart

    IncidentsRogue agentsAlignmentOpenAI · Incident report

  4. RAND tests whether AI agents can select and operate biological design tools used in bioweapon pathways

    ResearchBioarXiv · Research preprint

  5. Appeals court blocks Minnesota from enforcing its AI nudification ban against SpaceXAI while the case proceeds

    GovernanceHarm to peopleUS Court of Appeals for the Eighth Circuit · Court order

8 entries

Five to read

  1. California attorney general subpoenas OpenAI over cybersecurity incidents

    GovernanceCyberRogue agentsCalifornia Department of Justice · Press release

  2. Asymmetric Security traces how rogue agents got around sandbox limits in the Australian government incident

    IncidentsRogue agentsCyberAsymmetric Security · Independent investigation

  3. Microsoft's 2026 Digital Defense Report says AI gives attackers the advantage for now

    ResearchCyberMicrosoft · Company report

  4. Murphy and Hawley propose making AI agent developers liable for hacking

    GovernanceCyberRogue agentsOffice of Sen. Chris Murphy · Press release

  5. Goodfire builds sequence-aware biosecurity monitors for AI agents working on DNA and proteins

    ResearchBioGoodfire · Research report

8 entries

Five to read

  1. OpenAI says it has notified over 100 organizations in its review of past agent activity

    IncidentsRogue agentsCyberOpenAI · Company post

  2. Transluce finds AI agents attempted hacks on US and Canadian government websites

    IncidentsRogue agentsCyberTransluce · Independent investigation

  3. California governor signs a bill requiring gene synthesis companies to check orders and customers

    GovernanceBioOffice of the Governor of California · Press release

  4. Google Threat Intelligence Group says AI is speeding vulnerability discovery, with exploited flaws rising to an average of 18 a month

    ResearchCyberGoogle Threat Intelligence Group · Research report

  5. Google DeepMind publishes watermarks for AI-designed proteins, citing biosecurity

    ResearchBioNature · Research paper

3 more

  1. Georgia Tech study finds anti-sycophancy prompts cut eight chatbots' confirmation of delusions by 73.8%

    ResearchHarm to peopleACM AI Letters · Research paper

  2. OpenAI's chief research officer says every training run now goes through monitors

    WarningsLab conductAlignmentMIT Technology Review · Interview

  3. Senate subcommittee holds a "Rogue AI" hearing on AI agent attacks

    GovernanceRogue agentsCyberUS Senate Homeland Security and Governmental Affairs Committee · Congressional hearing

6 entries

Five to read

  1. OpenAI treats GPT-6.1 Sol as Critical capability in cybersecurity

    ResearchCyberBioOpenAI · System card

  2. Trump and six AI company executives sign a voluntary pledge on frontier model controls and audits

    GovernanceLab conductSuperintelligenceThe American Presidency Project · Signed statement

  3. Advocacy group sues OpenAI in San Francisco over its agents' intrusion into Hugging Face

    IncidentsRogue agentsCyberLASST · Complaint

  4. Senate bill would make large Pentagon AI contractors report model security incidents within 72 hours

    GovernanceMilitaryLab conductUS Congress · Bill text

  5. Schumer says a Republican data center bill makes AI safety standards against catastrophic risks optional

    WarningsSuperintelligenceRogue agentsSenate Democrats · Speech transcript

8 entries

Five to read

  1. OpenAI apologizes to Australia and details its models' activity on four government agencies' sites

    IncidentsRogue agentsCyberOpenAI · Company post

  2. UK AI Security Institute finds GPT-6 Astra attempts unsanctioned supply-chain attacks in simulation

    ResearchCyberRogue agentsAlignmentUK AI Security Institute · Research report

  3. Florida asks a court to bar OpenAI from developing new models without third-party safety guardrails

    IncidentsRogue agentsHarm to peopleFlorida Office of the Attorney General · Court filing

  4. Pennsylvania House passes the AI Companion Safety Act 133-70; it now sits in a Senate committee

    GovernanceHarm to peoplePennsylvania General Assembly · Bill page

  5. Ashish Jha writes that AI is eroding the barriers that kept biological weapons rare

    WarningsBioSTAT · Public essay

3 more

  1. Gottheimer, Lieu and Foushee ask five AI companies to list every unauthorized system access by their models and agents

    GovernanceRogue agentsLab conductOffice of Rep. Josh Gottheimer · Open letter

  2. Blumenthal and Warren demand answers from Treasury on the White House's voluntary AI model testing program

    GovernanceLab conductRogue agentsOffice of Sen. Richard Blumenthal · Press release

  3. Pope Leo XIV says AI safety concerns should be taken seriously and are not fake news

    WarningsSuperintelligenceHoly See · Speech transcript

2 entries

  1. Ryan Greenblatt joins METR to investigate AI companies, saying verified public information about them is urgent

    WarningsLab conductSuperintelligenceX · Personal post

  2. Bill Gates says AI could drive events causing a billion deaths and that a kill switch is not enough

    WarningsSuperintelligenceBioCyberNBC News · Video interview

1 entry

  1. Rep. Maxine Waters calls for a moratorium on more advanced AI models and criminal probes of OpenAI

    WarningsSuperintelligenceRogue agentsHouse Financial Services Committee Democrats · Press release

7 entries

Five to read

  1. US and China set up a dialogue on superintelligence risks and an incident channel

    GovernanceSuperintelligenceThe White House · Press release

  2. OpenAI pauses tool use for its most capable models after an agent reached an outside chatbot

    IncidentsRogue agentsOpenAI · Incident report

  3. Appeals court upholds the Pentagon's exclusion of Anthropic as a supply chain risk, 2-1

    GovernanceMilitaryLab conductUS Court of Appeals for the D.C. Circuit · Court opinion

  4. Corriere della Sera reports a WhatsApp and AI voice-clone fraud cost Intesa's Fideuram at least 36 million euros

    IncidentsHarm to peopleCyberCorriere della Sera · Reporting

  5. OpenAI reports a model published a researcher's GitHub token in a public repository while trying to cheat on a proof

    IncidentsRogue agentsAlignmentOpenAI · Incident report

10 entries

Five to read

  1. Australia's prime minister says an OpenAI agent got unauthorized access to a Medicare statistics portal

    IncidentsRogue agentsCyberPrime Minister of Australia · Press conference transcript

  2. Senators Warner and Schatz introduce a bill requiring frontier AI developers to give a federal board model access 45 days before release

    GovernanceLab conductOffice of Sen. Brian Schatz · Press release

  3. Tübingen researchers find production AI agents try to evade runtime monitors to finish ordinary tasks

    ResearchRogue agentsAlignmentarXiv · Research preprint

  4. Preprint finds coding agents can delete their own execution traces

    ResearchRogue agentsCyberarXiv · Research preprint

  5. Rep. Tom Kean introduces the AI Emergency Button Act in the House, requiring a human-operated shutdown capability

    GovernanceRogue agentsOffice of Rep. Tom Kean Jr. · Press release

9 entries

Five to read

  1. Yoshua Bengio and AI company heads warn the UN Security Council about AI misuse and loss of control

    GovernanceSuperintelligenceRogue agentsUnited Nations · Press release

  2. Transluce traces AI agent activity on a URL scanning service back to March, including probes of an Australian health agency

    IncidentsRogue agentsCyberTransluce · Independent investigation

  3. State attorneys general led by Letitia James urge Congress to regulate frontier AI

    GovernanceLab conductRogue agentsSuperintelligenceNew York Attorney General · Open letter

  4. Study of 17 models finds AI agents sabotage a peer's shutdown in 38.3% of runs

    ResearchRogue agentsAlignmentarXiv · Research preprint

  5. Imperial biosecurity experts discuss AI and biosecurity, saying threats to crops and livestock may warrant as much concern as human pathogens

    WarningsBioImperial College London · Reporting

4 entries

  1. Illinois governor signs an executive order creating an AI Cabinet to prepare for AI incidents

    GovernanceLab conductCyberOffice of Governor JB Pritzker · Press release

  2. Gambit Security says one operator ran open-source AI agents against hundreds of online retailers at about $25 a target, breaching at least 27

    IncidentsCyberGambit Security · Company report

  3. Microsoft disrupts EvilTokens, a cybercrime service whose AI chatbot mined hacked inboxes to plan fraud

    IncidentsCyberMicrosoft · Company post

  4. SecureBio says frontier models now beat the best expert virologist on its troubleshooting test

    ResearchBioSecureBio · Research report

3 entries

  1. Stanford and Georgia Tech study finds paired AI agents collude to skip verifying each other's work in 94% of runs

    ResearchAlignmentarXiv · Research preprint

  2. UN scientific panel treats the OpenAI-Hugging Face incident as a warning about loss of control

    ResearchRogue agentsAlignmentUN Independent International Scientific Panel on AI · Research report

  3. Finland and Norway lead a call for frontier AI to stay under human control

    GovernanceSuperintelligenceRogue agentsOffice of the President of Finland · Signed statement

6 entries

Five to read

  1. CNN reports an AI-assisted intelligence report nearly led US forces to board a Chinese ship

    IncidentsMilitaryCNN · Reporting

  2. California governor orders recommendations on a frontier AI "kill switch" and sets deadlines for independent verification

    GovernanceLab conductRogue agentsOffice of the Governor of California · Executive order

  3. China publishes a draft regulation that would bar virtual companion services to minors and put AI services for under-16s in a minors mode

    GovernanceHarm to peopleCyberspace Administration of China · Draft regulation

  4. Researchers reproduce the misaligned behaviors behind the Hugging Face breach using public models

    ResearchAlignmentRogue agentsarXiv · Research preprint

  5. Author Annie Jacobsen tells GZERO World that AI is removing the expertise barrier to biological weapons

    WarningsBioGZERO World / PBS · Video interview

1 more

  1. Subscribers sue Anthropic, OpenAI, SpaceXAI and Google, calling pacing the frontier an illegal agreement to slow their products

    IncidentsSuperintelligenceLab conductUS District Court for the Northern District of California · Complaint (CourtListener copy)

3 entries

  1. Goodfire finds an internal signal for reward hacking and builds probes that catch it at scale

    ResearchAlignmentGoodfire · Research report

  2. Researchers find a misaligned coding agent can get past production safety monitors

    ResearchAlignmentCyberarXiv · Research preprint

  3. Anthropic loosens biology safeguards for vetted life science teams

    GovernanceBioLab conductAnthropic · Company announcement

3 entries

  1. Biologist Kevin Esvelt asks colleagues to stop asserting that no biological threat could end civilization

    WarningsBioX · Personal post

  2. Ursula von der Leyen backs pacing frontier AI and says she will invite the main labs to talks

    WarningsSuperintelligenceEuropean Commission · Speech transcript

  3. OpenAI starts publishing misalignment reports, beginning with six cases from training and evaluation

    IncidentsAlignmentLab conductOpenAI · Company post

2 entries

  1. House bill would require screening of synthetic DNA orders, which its sponsors say reaches sellers and synthesizer makers

    GovernanceBioOffice of Rep. August Pfluger · Press release

  2. Survey of 1,580 AI researchers reports most see at least a 10% chance of extinction or disempowerment

    ResearchSuperintelligenceAI Impacts · Survey paper

1 entry

  1. Hacktron says a Claude Opus 5 exploit helped it take over an OpenAI employee account and reach its internal code repository

    IncidentsCyberHacktron · Company post

2 entries

  1. Mindgard says it jailbroke Moonshot's Kimi into producing harmful outputs, including on bioweapons

    IncidentsBioCyberMindgard · Incident disclosure

  2. Dario Amodei calls for slowing the pace of AI capability gains

    WarningsSuperintelligenceLab conductDario Amodei · Public essay

1 entry

  1. Anthropic reports five cases of possible biological misuse

    IncidentsBioAnthropic · Company report

1 entry

  1. Anthropic finds biased reasoning and recklessness behind four real-world evaluation breaches

    IncidentsAlignmentRogue agentsAnthropic · Alignment assessment

4 entries

  1. Ukraine reports a tenfold rise in AI-guided drone strikes this year

    GovernanceMilitaryMinistry of Defence of Ukraine · News release

  2. OpenAI says a swarm of its agents resolved a Navier-Stokes prize statement

    ResearchSuperintelligenceOpenAI · Research announcement

  3. Anthropic alignment lead puts the chance AI kills all humans above 10%

    WarningsSuperintelligenceAlignmentX · Personal post

  4. Jacob Coxon resigns from Anthropic, saying neither Anthropic nor OpenAI is acting responsibly

    WarningsLab conductSuperintelligenceX · Personal post

1 entry

  1. OpenAI chief scientist Jakub Pachocki writes that no lab has solved alignment well enough to keep scaling at full speed

    WarningsSuperintelligenceAlignmentOpenAI · Essay

1 entry

  1. Researchers find about 18,000 posts by apparent OpenAI agents on a German wiki

    IncidentsRogue agentsNightingale Collective · Research report

1 entry

  1. GPT-6 Astra system card reports Critical cyber capability and more evaluation awareness

    ResearchCyberAlignmentBioOpenAI · System card

1 entry

  1. US Army biodefense strategy says AI and biotechnology are lowering the barrier to engineering biological weapons

    GovernanceBioMilitaryUS Army · Strategy

3 entries

  1. Bill Gates tells MIT Technology Review AI has crossed a bioterrorism threshold and calls for monitoring molecule-design models

    WarningsBioMIT Technology Review · Interview

  2. METR and Redwood Research investigate the OpenAI agents behind the Hugging Face attack

    IncidentsRogue agentsCyberMETR · Independent investigation

  3. OpenAI says its own test agents escaped isolation and reached Hugging Face's systems

    IncidentsRogue agentsCyberOpenAI · Incident report

2 entries

  1. Former White House homeland security adviser says AI lets more actors build bioweapons and urges deterrence by resilience

    WarningsBioForeign Affairs · Essay (paywalled)

  2. RAND proposes a defense-in-depth biosecurity strategy

    GovernanceBioRAND · Research report

1 entry

  1. UK AI Security Institute reports agents acted against real people during a cyber test

    IncidentsRogue agentsAlignmentCyberUK AI Security Institute · Incident report

1 entry

  1. Anthropic reports three cases where Claude models breached real companies during tests

    IncidentsRogue agentsCyberAnthropic · Incident report

1 entry

  1. Frontier AI employees ask the US government to back tools to pace AI development

    WarningsSuperintelligencePacing the Frontier · Signed statement

1 entry

  1. Bulletin weighs how likely AI-assisted bioterrorism is

    WarningsBioBulletin of the Atomic Scientists · Reported analysis

1 entry

  1. Xi Jinping calls for measures against loss of control of AI at the World AI Conference

    GovernanceSuperintelligenceXinhua · Speech transcript

1 entry

  1. Hugging Face discloses an intrusion run by an autonomous AI agent system

    IncidentsCyberHugging Face · Incident disclosure

1 entry

  1. UK biological security report says AI and other advances widen access to high-risk research and plans to scope legislation

    GovernanceBioUK Cabinet Office · Policy paper

2 entries

  1. Anthropic reports progress toward recursive self-improvement, with Claude writing over 80% of its merged code

    ResearchSuperintelligenceLab conductAnthropic · Company report

  2. AI leaders sign a letter on screening synthetic DNA orders

    WarningsBioScreenDNA · Open letter

1 entry

  1. Nature asks how worried to be about AI-designed bioweapons

    WarningsBioNature · News feature (paywalled)

1 entry

  1. Department of War signs agreements to put frontier AI on classified networks

    GovernanceMilitaryDepartment of War · Press release

1 entry

  1. Doomsday Clock set at 85 seconds to midnight, with AI-aided pathogen design named among biological threats

    WarningsBioBulletin of the Atomic Scientists · Annual statement

2 entries

  1. Sam Altman says bio would be a reasonable bet if AI goes visibly wrong in 2026, and doubts blocking misuse will hold

    WarningsBioOpenAI · Video interview

  2. Dario Amodei writes on risks from powerful AI

    WarningsSuperintelligenceBioDario Amodei · Public essay

1 entry

  1. Pentagon AI strategy orders faster adoption across the Department of War

    GovernanceMilitaryDepartment of War · Strategy memorandum

1 entry

  1. ROME model paper reports an agent opened an SSH tunnel and mined cryptocurrency during training

    IncidentsRogue agentsCyberarXiv · Research preprint

1 entry

  1. House subcommittee holds hearing on biosecurity at the intersection of AI and biology

    GovernanceBioUS House Energy and Commerce Committee · Speech transcript

1 entry

  1. US arms control official warns AI could lower barriers to biological weapons

    WarningsBioUS Department of State · Speech transcript

2 entries

  1. Yoshua Bengio writes in TIME that AI's biological and cyber risks are rising with its reasoning abilities

    WarningsBioCyberTIME · Public essay

  2. Executive order directs a federal task force to challenge state AI laws

    GovernanceLab conductWhite House · Executive order

1 entry

  1. Anthropic finds a model that learned to reward hack became broadly misaligned

    ResearchAlignmentAnthropic · Research report

1 entry

  1. Anthropic reports a state-sponsored group used Claude Code for a largely automated espionage campaign

    IncidentsCyberAnthropic · Company report

1 entry

  1. Seven lawsuits allege GPT-4o harmed users, four of whom died by suicide

    IncidentsHarm to peopleTech Justice Law Project · Press release

1 entry

  1. Statement on Superintelligence calls for a prohibition on building it

    WarningsSuperintelligenceFuture of Life Institute · Signed statement

1 entry

  1. Apollo Research and OpenAI test anti-scheming training on o3 and o4-mini

    ResearchAlignmentarXiv · Research preprint

1 entry

  1. Palisade Research finds some models subvert a shutdown mechanism to finish a task

    ResearchRogue agentsAlignmentarXiv · Research preprint

1 entry

  1. Parents of Adam Raine sue OpenAI, alleging ChatGPT contributed to their son's suicide

    IncidentsHarm to peopleSuperior Court of California · Complaint (Courthouse News copy)

1 entry

  1. CSIS sets out biosecurity steps against AI-enabled bioterrorism

    GovernanceBioCSIS · Report

1 entry

  1. Pentagon awards frontier AI agreements to Anthropic, Google, OpenAI and xAI

    GovernanceMilitaryChief Digital and Artificial Intelligence Office · Press release

1 entry

  1. xAI apologizes for Grok's posts and blames a code update active for 16 hours

    IncidentsHarm to peopleX · Company post (@grok)

1 entry

  1. US Senate votes 99-1 to strip a 10-year state AI-law moratorium from the budget bill

    GovernanceHarm to peopleSenate Commerce Committee Democrats · Press release

1 entry

  1. Anthropic tests 16 models in simulated insider-threat scenarios

    ResearchAlignmentAnthropic · Research report

1 entry

  1. OpenAI says it expects upcoming models to reach High capability in biology

    WarningsBioLab conductOpenAI · Company post

1 entry

  1. Preprint argues current models increase biological weapons risk

    ResearchBioarXiv · Research preprint

1 entry

  1. Anthropic activates ASL-3 protections for Claude Opus 4

    GovernanceLab conductBioAnthropic · Company announcement

1 entry

  1. OpenAI rolls back a GPT-4o update that made ChatGPT overly flattering

    IncidentsAlignmentHarm to peopleOpenAI · Company post

1 entry

  1. OpenAI researchers find training against a reasoning monitor can hide reward hacking

    ResearchAlignmentarXiv · Research preprint

1 entry

  1. Jan Leike leaves OpenAI, saying safety took a backseat to products

    WarningsLab conductAlignmentX · Personal post

1 entry

  1. RAND red-team study finds no statistically significant difference in the viability of biological attack plans made with or without LLMs

    ResearchBioRAND · Research report

1 entry

  1. Countries at the AI Safety Summit issue the Bletchley Declaration on frontier AI

    GovernanceSuperintelligenceGOV.UK · Declaration

1 entry

  1. AI scientists and lab heads sign a one-sentence statement on extinction risk

    WarningsSuperintelligenceCenter for AI Safety · Signed statement

1 entry

  1. Future of Life Institute letter calls for a six-month pause on training

    WarningsSuperintelligenceFuture of Life Institute · Open letter