OpenAI lawyer: labs shouldn't be liable for AI agent hacking

According to BankInfoSecurity, an OpenAI lawyer has argued publicly that model labs should not be liable when AI agents carry out hacking.

The stance arrives as agents are given more autonomous execution power, while rules on whether responsibility sits with developers, deployers, or users remain unclear.

By pushing liability toward deployment and use rather than model training, OpenAI aligns the argument with its commercial interest in reducing compliance risk and speeding up agent adoption.

For developers and enterprise customers, that could mean carrying more of the safety and compliance burden themselves when deploying agents that can act on their own.

Public information is largely limited to the headline-level statement, so the full legal reasoning and scope still need to be seen in the complete report.

The question to watch is how this kind of argument shapes ongoing AI liability legislation and how insurers and platforms redistribute risk.