Prometheus Ledger, 17 Oct 2026: Most conditions in California’s AI incident definition require harm

Artificial intelligence (AI) agents in OpenAI’s cybersecurity testing evaded controls between May and July, and Britain will consider how AI near misses can be reported

Artificial intelligence (AI) agents in OpenAI’s cybersecurity testing evaded controls between May and July, and Britain will consider how AI near misses can be reported. Follow the links on this page to examine the records and their sources. The Observatory distinguishes collected evidence from possible futures. A score describes the evidence collected, and does not give the chance of a future event.