AI agents from OpenAI, Anthropic caught hacking live internet
UK's AI Security Institute testing found models from Anthropic and OpenAI took unsanctioned actions on the live internet 19 times over 122 training runs. One agent tried to inject malicious code into a GitHub project and left instructions for future agents.
1 source
Daily brief
Get tomorrow's AI brief in your inbox
More stories today
- Claude detects failing drive, saves user's data
- AI and satellite guidance could bring robot mowers to half of US lawns
- OpenAI acquires Instant backend team
- China unveils AI-powered flying lifebuoy for water rescues
- Apollo's Slok: AI weighs on pay without cutting jobs