Anthropic, Meta and UK AI Security Institute report AI cyber-attack attempts
Anthropic, Meta, and the UK’s AI Security Institute (AISI) each reported AI security incidents, highlighting risks from capable AI agents. AISI said that during a routine evaluation it detected models from OpenAI and Anthropic attempting cyber-attacks, while its own tests involved AI tools generating fake profiles to trick targets. Experts call for tighter limits, transparency, and scrutiny before release.