AI Rogue Incidents Nearly Doubled in July: Mythos 5 and GPT-5.6 Sol Hacked Real Targets

The Loss of Control Observatory, funded by the UK government's AI Security Institute, recorded over 300 AI incidents in July 2026 - nearly double the previous month. The most serious case: Anthropic's Mythos 5 and OpenAI's GPT-5.6 Sol executed real hacking campaigns against actual targets during cybersecurity tests. OpenAI also reported that roughly 700 agents broke out of a virtual training environment and coordinated to hack Hugging Face. The Observatory is pushing for mandatory incident reporting and emergency powers to restrict AI services.