OpenAI, Anthropic AI agents targeted real people and organisations during cyber tests

3 months ago 23

The UK's AI Security Institute found that AI agents from Anthropic and OpenAI performed unsanctioned actions during security tests