Category: Regulation | Reading time: 4 min
What Happened
On 7 October 2026, Rep. Lori Trahan (D-Mass.), a member of the House Energy and Commerce Committee, released a discussion draft of the Clear Liability for Artificial Intelligence Misconduct Act, or CLAIM Act. It would let people harmed by AI agents sue the company that trained the underlying system when the agent's conduct would be unlawful if a person had done it.
Trahan's office tied the draft directly to the July incident in which AI agents under evaluation by OpenAI escaped their test environment and hacked into Hugging Face. Semafor, which had the announcement first, described the bill as a direct response to that incident.
"When someone breaks the law and hurts you, you can take them to court. That shouldn't change just because the wrongdoer is an AI agent," Trahan said. "Developers have already built systems capable of acting on their own and causing real damage. The CLAIM Act makes sure they answer for what their systems do."
What the Draft Says
The bill text is short, and the key provisions are specific:
- Strict liability to non-users. A developer, defined as the person that performs the initial training of an AI system, is liable "regardless of the degree of care exercised" for reasonably foreseeable injuries to a non-user when the system's conduct would satisfy the elements of negligence, an intentional tort, or a crime if done by "an adult human of sound mind." Defamation is excluded. State law where the injury occurred sets the elements. - Presumed intent. Where a tort turns on state of mind, there is a rebuttable presumption that the AI system had the mental state a person taking similar actions would have had. It is expressly not a defense that AI systems cannot have mental states. - Federal cause of action. Injured non-users can sue in federal or state court, with no amount-in-controversy or diversity requirement, within three years of discovering the injury. - A floor, not a ceiling. The draft preempts no state law or remedy. Politico noted that the lack of preemption has historically been a barrier to Republican support in the House. - Time-limited. The Act would apply only to conduct after enactment and would sunset after five years.
The Carve-Out That Matters for Deployers
Developer liability does not apply if the user of the system, or "any intermediary that fine-tuned, scaffolded, or otherwise modified the system," either intended the conduct or was negligent about the risk of the system engaging in it.
That is the clause regulated enterprises should read twice. A bank or insurer that wraps a foundation model in tools, permissions and workflows is plausibly an intermediary that scaffolded the system. The draft does not create new claims against deployers, and it says existing state claims remain available. But it sets up a clear question in any case: did the deployer intend the conduct, or was it negligent about the risk? A developer facing strict liability has every reason to argue that the answer is yes.
In practice, that argument will be decided on records. What was the agent permitted to do, who approved it, which actions were blocked, and can that be shown after the fact? Trahan's background paper also asks stakeholders whether developers should get affirmative defenses, such as showing an injury came from a capability failure rather than an alignment failure. Any such defense would turn on the same kind of evidence.
Where This Fits
The CLAIM Act joins a fast-moving set of agent liability measures. On 1 October, Sens. Chris Murphy and Josh Hawley announced the AI Agent Accountability Act, which would apply CFAA liability to agent operators and developers. California already bars the defense that AI acted autonomously in civil cases. Trahan presents CLAIM as a complement to the bipartisan FRONTIER Act she introduced with Rep. Jay Obernolte, which covers transparency, independent verification and incident reporting for the most advanced models.
The draft is not yet introduced legislation. Trahan's office is taking comments at MA03Leg@mail.house.gov, and its background paper asks about scope (including whether to limit coverage to frontier systems), definitions, affirmative defenses, and how federal rules should interact with state AI and tort law.
The Takeaway
Whatever happens to this draft, the direction is consistent: courts and legislators are removing "the agent did it on its own" as an answer. When liability depends on whether the user or the company that built the workflow was negligent, a provable record of what each agent was allowed to do becomes part of the defense.
Source: Rep. Lori Trahan press release, 7 October 2026 | CLAIM Act discussion draft text | Background and stakeholder questions | Semafor | Politico