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

2 months ago 22

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