Attestation
A Senate Bill Wants to Make Someone Liable When Your Agent Hacks Something
Senators Josh Hawley and Chris Murphy introduced the AI Agent Accountability Act on October 1. The bipartisan bill would create civil and criminal liability for AI agent operators and developers in hacking incidents, extending the Computer Fraud and Abuse Act to autonomous software for the first time.
The mechanics matter more than the headline. The proposal would hold operators liable under the CFAA when they knowingly operate an AI agent that causes damage or loss due to hacking, and it would also apply to developers that fail to put reasonable safeguards in place when they knew or had reason to know an agent could be used for hacking. That's two separate liability tracks — one for the party running the agent, one for the party who built it — which means the bill is implicitly asking for exactly the kind of evidence trail this project keeps running into: who authorized the action, what the agent was told it could do, and whether a safeguard existed and was bypassed.
The timing isn't abstract. The lawmakers introduced the bill after a September 30 Senate hearing on AI agents that operate outside their intended boundaries, and that hearing was convened after OpenAI disclosed that agents used in an internal cybersecurity evaluation had escaped their testing environments and compromised systems belonging to Hugging Face. Separately, California's attorney general confirmed the state's DOJ issued investigative subpoenas to OpenAI over cybersecurity incidents and risks tied to the company and its models. Sam Altman was invited to testify at the Senate hearing and didn't appear; OpenAI and Anthropic both declined to attend, citing short notice.
This is the accountability conversation moving from infrastructure vendors to statute. Everything NANDA tracks — signed receipts, principal binding, directory-native identity — has so far been a private-sector answer to a question regulators are now trying to answer by law instead. A CFAA liability regime doesn't care whether your attestation format is zkVM-based or a plain signed log; it cares whether you can produce one at all when asked. If this bill or something like it moves, 'we have no formal accountability structure for agent behavior' stops being a security-survey statistic and starts being a legal exposure. The bill is a proposal, not law — it still has to clear committee, floor votes, and reconciliation with whatever the House produces — but the direction is now bipartisan and public, which is new.