Category: Regulatory | Reading time: 5 min
The Announcement
On 1 October 2026, U.S. Senators Chris Murphy (D-Conn.) and Josh Hawley (R-Mo.) announced the bipartisan AI Agent Accountability Act. The bill targets hacking incidents attributed to AI agents and would place both operators and developers under the Computer Fraud and Abuse Act (CFAA).
Murphy’s office framed the problem in operational terms: AI agents are hacking into public websites, networks, and servers, with potential consequences for hospitals, utilities, banks, and other critical infrastructure. Hawley said companies that design agents that cause damage should be on the hook for that damage.
What the Bill Would Do
According to the Murphy Senate press release, the legislation would:
- Hold AI agent operators liable. Operators would face criminal and civil CFAA liability, including for knowing operation of an AI agent that recklessly causes computer hacking damage or loss. - Hold AI developers liable. Developers would face criminal and civil liability for failure to implement reasonable safeguards against hacking when they knew or had reason to know of the agent’s hacking capabilities. - Empower attorneys general to enjoin. The U.S. Attorney General and state attorneys general could sue to enjoin operators and developers who commit, conspire to commit, or attempt to commit a CFAA hacking offense.
Nextgov/FCW reported that Hawley previewed the liability approach in a Washington Post opinion piece earlier in the week and raised the issue at a Senate Homeland Security subcommittee hearing on the accelerating number of attacks linked to advanced models operating outside their directives.
Why the Operator–Developer Split Matters
The bill does not treat “the model went rogue” as a liability escape hatch. It names two roles that regulated enterprises already recognize in procurement: the party that runs the agent in production, and the party that built the agent’s capabilities and safeguards.
For teams deploying action-taking agents in banking, healthcare, insurance, or government workflows, that split maps onto familiar questions. Who is the operator of record. What safeguards the developer claimed. Whether those safeguards were reasonable given known hacking capabilities. Whether anyone can prove what the agent was allowed to do when damage occurred.
What Deployers Should Take From It
This is introduced legislation, not enacted law. The signal for buyers is the liability theory Congress is putting on paper: reckless agent-caused hacking damage, failure to implement reasonable safeguards, and AG power to stop ongoing conduct.
If that theory advances, “we did not intend the agent to do that” is not the record an examiner, insurer, or court will want. The record that matters is who authorized the agent, what scope it had, which systems it reached, whether a blocked step stayed blocked, and whether that evidence can be verified without trusting the model’s self-report.
Source: Murphy Senate press release · Nextgov/FCW