World Internet Conference report finds fewer than 30% of 24 countries have relatively stable arrangements for frontier AI incident reporting
World Internet Conference · Policy report
Read the source (opens in a new tab)- What happened
- The World Internet Conference's Specialized Committee on AI published "Frontier AI Safety and Governance 2026" on November 3, 2026, a report co-led by Zeng Yi and Seán Ó hÉigeartaigh.
- Key points
- It says misuse and criminal, information integrity and critical system risks "have already translated into tangible real-world harms to a considerable extent."
- Of 24 countries assessed on public information, most are better prepared in institutions than in technical safety and evaluation, verification and compliance, and accountability.
- Fewer than 30% have relatively stable arrangements for deployment control, supply chain and open-source governance, or incident and near-miss reporting.
- Where harm may be widespread and hard to reverse, it calls for pre-deployment dangerous capability evaluations, preset thresholds, and options for "human takeover, reduced functionality, rollback and shutdown."
- Evidence
- Policy report from an international organization's AI committee, drawing on published safety evaluations, recorded AI incidents and expert assessments, plus a desk review of 24 countries' public arrangements.
- Uncertain
- The report says its country sample is not a ranking of any country's governance capacity. Readiness findings rest on public information up to August 31, 2026.