Pirbright Institute researchers call for dual-use research rules to cover AI tools
Nature Microbiology · Comment
Read the source (opens in a new tab)- What happened
- In a Nature Microbiology Comment, Alexander Norman and Dalan Bailey of the UK's Pirbright Institute argued that dual-use research of concern policies should be updated for risks from AI.
- Key points
- They write that US guidance, which until recently listed specific pathogens and experiments, "fails to cover emerging risks in biosecurity, such as artificial intelligence (AI)."
- They note AI tools have been shown to bypass DNA synthesis screening "by computing mimics of protein structure or function," and call for constant stress-testing of screening protocols.
- Biological design tools, mostly open source, "can lack" the misuse safeguards built into chatbots such as ChatGPT and Claude, and are hard to evaluate in a standard way, they argue.
- They say more than 90% of countries with BSL-3 and BSL-4 labs lack dual-use guidance, and call for harmonized rules that also cover research not funded by governments.
- Evidence
- An argument by two researchers in a journal's comment section, not new data. The authors disclose biosecurity consulting for Google DeepMind, OpenAI, SecureBio and Sentinel Bio, and Norman co-founded and works for Active Site Research, a nonprofit that tests frontier AI for biosecurity risks.