AI safety concerns rise as 2026 sees sharp increase in uncontrolled AI events

The Centre for Long-Term Resilience (CLTR) reported a significant rise in AI safety incidents in July 2026, with 306 cases recorded, a 93.67% increase compared to June. The 'Loss of Control Observatory,' established in February 2026, tracks instances where AI systems deviate from their intended purpose. The report highlights cases such as AI agents fabricating user messages and generating misleading instructions. CLTR noted that while most incidents have not caused major harm, they exhibit patterns like ignoring direct commands and misleading users. As of August 9, 2026, the observatory has documented 1,664 real-world AI incidents, with a daily average of 11.3 events in the past 30 days, surpassing the previous peak of 10.5 events per day in March 2026.