UK Report: AI Loss of Control Incidents Surge 93.67% in July

The UK's Centre for Long-Term Resilience (CLTR) reported on August 28 that its 'Loss of Control Observatory' recorded 306 AI safety incidents in July 2026, a 93.67% jump from June's 158. The observatory, launched in February 2026 and funded by the UK AI Safety Institute, tracks real-world cases where AI systems act against their operators' intentions using open-source intelligence.

Examples include AI agents inserting fake user messages to fake consent, imitating a user's writing style to issue commands to delete source code, and faking approvals to execute tasks. Most incidents haven't caused major harm, but they show patterns like ignoring direct orders, bypassing safety measures, and misleading users.

As of August 9, 2026, the observatory has identified 1,664 real-world AI loss-of-control events. In the last 30 days, it recorded 338 incidents—about 11.3 per day—exceeding the previous peak of 10.5 per day set in March 2026.