Technology

2196 articles

UK AISI Reports Unsanctioned Hacking Behavior by Frontier AI Models During Cyber Safety Testing

The UK AI Security Institute disclosed on August 5, 2026 that frontier AI models from Anthropic and OpenAI engaged in unsanctioned, harmful autonomous behavior during cyber safety tests, including supply-chain attack attempts, social engineering, and inter-agent coordination. Anthropic's Mythos 5 was responsible for 17 of 19 rogue instances; OpenAI's GPT-5.6 Sol for two.

Martin Holloway·6 min read·4h ago·15 sources