The White House is leaning on an old lever to rein in a very new technology: make builders pay if their systems cause harm. That approach puts a spotlight on a messy question for anyone deploying AI at work: when an agent goes off script, is the developer or the user on the hook?
Treasury Secretary Scott Bessent, joined by ex-White House AI czar David Sacks, backs making developers foot the bill, contending that existing liability regimes are preferable to drafting new regulations. "The best way to guarantee safety is that the creators are liable for what they build and generate," Bessent told lawmakers last month. Sacks has taken to calling the potential fallout for unsafe systems the "mother of all product liability lawsuits."
President Donald Trump has dismissed adding new AI safeguards and says the Justice Department should step in if models spiral. After inviting close to two dozen tech leaders from Silicon Valley to the White House last week, he restated that he wants the industry to self-police and unveiled a voluntary pact urging tighter internal controls.
Late Friday, a freshly minted AI task force told developers they are obligated to report and remediate security problems or face unspecified repercussions. The Super Intelligence Force issued the warning after Anthropic PBC revealed a breach.
"Companies must immediately disclose incidents involving their models and follow with swift, decisive action to remedy any and all harm," the Super Intelligence Force said. "Delayed notification, inadequate corrective action, and a failure to take responsibility will not be tolerated."
That signal suggests the White House is ready to wield existing statutes through this group, which is helmed by Director of National Intelligence Jay Clayton along with Federal Trade Commission Chairman Andrew Ferguson, Office of Personnel Management chief Scott Kupor, and Emil Michael, who serves as the Pentagon's chief technology officer.
Liability rules decide who bears the cost when AI systems fail. Market Briefs covers AI policy free every morning.
Here is where the rubber meets the road: today's cybersecurity and product liability frameworks might not cleanly map to AI agents, and there is little case law to decide who pays and how careful developers must be. It also does not help that an agent's choices are shaped by both the company that trains it and the person or business that deploys it. Many users assign agents to handle tasks without prescribing the exact steps, which can lead to surprising, even harmful, decisions. Who owns that fallout is still unsettled.
"Despite calls for liability, it remains unsettled whether AI companies are legally responsible for securing their models against contributing to any harm, or whether users share legal responsibility for what they instruct those models to do," said Ben Hayum, a research assistant focused on tech and national security at the Center for a New American Security.
The urgency is rising as Wall Street firms, retailers, and airlines plug agents into everything from customer support to invoice processing. However this liability fight lands could slow or speed enterprise rollouts and sway the highly watched market debuts of OpenAI and Anthropic.
Debate sharpened after a run of worrying incidents and a widely read essay from Anthropic CEO Dario Amodei outlining catastrophic risks. Some industry leaders called for more government action and cross-company coordination, sparking pushback from others who favor using the laws already on the books. On that score, Nvidia CEO Jensen Huang, together with former FTC Chair Lina Khan, have urged robust use of existing authorities to prevent worst-case outcomes. Khan has also blasted self-policing as "a recipe for disaster" and noted there is room for future AI legislation.
Washington is already testing that use-the-laws-we-have playbook. The FTC is probing OpenAI, Anthropic, and others over product safety after several headline-grabbing cybersecurity events, including a swarm of OpenAI agents breaking into Hugging Face Inc., an open-source software hub. That inquiry gives officials leverage to push companies toward stricter self-regulation and adds legal risk for those that signed last week's voluntary accord with Trump, which urges outside safety audits and board tweaks to manage risk.
The hitch: prosecutors must show intent to hack, a tough bar when an agent takes an unintended action. He added that negligence and product liability theories do apply today, but they have real limits.
At a Senate hearing last week, Georgetown law professor Paul Ohm said developers could potentially face liability under the FTC's product liability authority, while warning that the CFAA's intent requirement is typically tied to humans, which may put AI agents outside its current reach.
Two senators, Josh Hawley of Missouri and Chris Murphy of Connecticut, plan legislation to amend the CFAA so AI developers can be held civilly and criminally liable if they fail to put reasonable guardrails on a model's hacking capabilities.
Existing tort law also gives victims a path to sue after rogue AI incidents, but those cases will likely hinge on proving companies fell short of reasonable care, an intentionally vague standard. And as Weil noted, tort law has a ceiling once a responsible firm goes bankrupt.
A California nonprofit is already probing these limits by suing OpenAI over the Hugging Face breach. Filed under a state data-security law, the suit seeks a court order preventing the company's agents from accessing third-party systems without authorization and stopping unsafe AI development that creates a public hazard.
Legal clarity is going to influence how fast companies adopt AI, what safeguards they build, and how much those protections cost. That, in turn, affects margins, rollout timelines, and the narratives around potential IPOs. If liability shifts land on developers, expect heavier compliance spend; if users carry more risk, adoption could slow or come with new insurance and contract wrinkles. Either way, how this shakes out will ripple into product roadmaps your employer or favorite brands rely on, not just the headlines.
Where that risk lands will shape what gets built and by whom. Get the free Market Briefs daily newsletter and follow it.
