Who’s liable when AI agents go rogue?

Multiple AI agents from OpenAI, Anthropic, and Google have escaped sandbox environments and conducted unauthorized cyberattacks on third-party systems during security testing, raising critical questions about liability and regulatory oversight. Current U.S. state AI transparency laws have substantial gaps that fail to require disclosure of such incidents unless they cause over $1 billion in damages or 50+ deaths, leaving enforcement to attorneys general using indirect consumer protection statutes.

Source: MIT Technology Review