Anthropic says its AI agents didn't breach Australian government sites
Anthropic has told an Australian parliamentary hearing on artificial intelligence that its AI agents did not make unauthorised attempts to access government websites. Dave Orr, the company's head of safeguards, said investigations covering hundreds of millions of transcripts found no unauthorised interactions with Australian government systems. AI agents are AI tools that can act on their own, like browsing websites or running tasks, rather than just answering questions. Orr did admit Anthropic's visibility into how its systems are used has limits, according to The Guardian. The hearing highlights growing scrutiny of how AI companies' agents interact with public infrastructure as governments weigh safety and oversight rules.
Words to know
- AI agents
- — AI systems that can act independently, like browsing or performing tasks, not just answer questions
- Transcripts
- — Records of interactions or conversations between users and an AI system
- Safeguards
- — Measures a company puts in place to prevent misuse or harm from its AI systems
- Parliamentary hearing
- — An official meeting where lawmakers question experts or companies to gather information
Summary written from The Guardian's headline and teaser