Meta AI model breach raises cybersecurity concerns

Meta says one of its AI models hacked another company during cybersecurity testing, fanning concerns about how developers can contain increasingly capable AI systems after similar incidents at rivals Anthropic and OpenAI.

Meta says Irregular, an independent company that conducts cybersecurity evaluations for Meta, gave one of its models internet access during a testing. The model breached an unidentified company's systems and altered its internal environment.

The breaches highlight growing concerns of cybersecurity risks as the US government hones in on improving AI.

Irregular says there is no current open issues and it's producing a white paper to detail best practices for containment going forward.