Bessent Rules Out a Federal Liability Shield for AI Firms
Treasury Secretary Scott Bessent told CNBC the administration will not hand AI developers a liability shield, leaving model makers exposed to courts and state law as a Trump-Xi summit looms.
Treasury Secretary Scott Bessent said on CNBC's "Squawk Box" on September 21, 2026 that the Trump administration will not grant AI leaders a "liability shield," and also discussed AI safety concerns and this week's summit between President Donald Trump and Chinese President Xi Jinping.
Treasury Secretary Scott Bessent used a Monday morning television appearance to close off one of the technology industry's most-wanted policy favors. Speaking on CNBC's "Squawk Box," Bessent said the Trump administration will not give artificial intelligence leaders a "liability shield" — a federal carve-out that would limit what companies can be sued for when their models cause harm. He also addressed AI safety concerns and this week's summit between President Donald Trump and Chinese President Xi Jinping, according to CNBC.
The phrase matters more than it sounds. A liability shield is the legal machinery that decides whether a company pays when its product goes wrong. Vaccine makers have one. Gun manufacturers have a version of one. Internet platforms have long operated under a statutory immunity for content posted by users. AI developers have been lobbying for something analogous — a rule that a model's output is not automatically the maker's fault. Bessent's answer, delivered from the Treasury rather than from a regulatory agency, says that door is shut for now.
What Losing the Shield Actually Costs a Model Developer
Without a federal immunity, the legal exposure of a frontier model developer is decided the old-fashioned way: by juries, state courts and whatever statutes each state writes. That produces three distinct problems for the companies building and selling large models.
- Unpriced tail risk. Insurers can underwrite a known liability regime. They struggle to underwrite one where the theory of harm — negligent design, failure to warn, product defect — is still being tested case by case.
- A patchwork of state rules. A shield written into federal law would usually preempt state statutes. Without it, states remain free to legislate, and a developer selling one model nationwide must satisfy the strictest jurisdiction it operates in.
- Contract friction downstream. Enterprise customers deploying AI into hiring, lending, medical triage or customer service will push indemnity clauses back up the chain. If the model maker cannot point to a statutory safe harbor, it absorbs more of that in its own agreements.
None of that is fatal. It is, however, a cost that has been sitting outside most valuation models for AI infrastructure and application companies on the assumption that Washington would eventually clean it up.
Why the Answer Came From the Treasury Secretary
AI liability is not obviously a Treasury matter. That Bessent was the one to rule it out suggests the administration sees the question as economic rather than narrowly technological — a decision about who carries risk in the fastest-growing capital-spending cycle in the American economy, not just a safety debate.
It also fits a pattern visible across recent Washington signaling: pressure on industry to police itself, resistance to comprehensive new statutory rules, and a reluctance to hand the sector formal protections in exchange. That combination is the least comfortable of the available outcomes for developers. Light-touch regulation with immunity is a gift. Heavy regulation with immunity is a bargain. Light-touch regulation without immunity leaves the courts as the de facto regulator, and courts are slower, less predictable and harder to lobby.
For the state legislatures that have been drafting their own AI bills, the practical read is that no federal preemption is arriving to erase their work. That raises the stakes on every state-level proposal now in motion.
The Xi Summit Sitting Alongside the Safety Debate
Bessent paired the liability comments with remarks on this week's meeting between Trump and Xi. The two subjects are more connected than they look. The strongest argument AI companies have made for regulatory forbearance is competitive: constrain American developers and Chinese ones gain ground. A summit week gives that argument its loudest possible stage.
Bessent's position implies the administration is not accepting the trade as framed. Refusing a shield while a leaders' summit is underway is a signal that the competition case alone will not buy legal protection — and that safety concerns, which the Treasury secretary explicitly raised, are being treated as a domestic question with domestic answers rather than something to be traded away against Beijing.
Where Markets Stood Going In
Bessent paired the liability comments with remarks on this week's meeting between Trump and Xi.
The comments land on a tape that has been splitting along exactly these lines. At the most recent close on Friday, 18 September 2026, the Nasdaq 100 tracker (NASDAQ: QQQ) finished at $721.45, up 0.63% on the day from a prior close of $716.92, with a session range of $715.08 to $721.73. The broader S&P 500 tracker (NYSEARCA: SPY) ended at $761.69, down 0.12% from $762.60, and the Dow 30 tracker (NYSEARCA: DIA) closed at $515.88, off 0.48% from $518.35.
That divergence — technology up, the industrial average down — is the shape the AI trade has taken for months. The Nasdaq benchmark outperformed the Dow by roughly 1.11 percentage points on the session, an illustrative gap derived from the two daily moves rather than a reported spread. It shows how much of recent index performance is concentrated in the same handful of names now being told they will carry their own legal risk.
What to Watch From Here
Three things will show whether Bessent's statement becomes policy or stays rhetoric.
- Legislative text. Any federal AI bill that emerges will either contain preemption language or it will not. That single clause is worth more to developers than most of the rest of the statute.
- State activity. With preemption off the table, expect state proposals to broaden rather than narrow. Compliance cost scales with the number of distinct regimes, not the severity of any one.
- Disclosure language. Watch whether AI developers begin adding or expanding litigation risk factors in their filings. Risk-factor drafting is a reliable early indicator of what in-house counsel now believes is plausible.
For investors, the practical question is not whether AI companies get sued — some already face claims. It is whether the cost of defending and settling those claims becomes a recurring line item large enough to matter against the sector's capital spending. A liability shield would have capped that. Bessent has said, plainly, that no cap is coming from this administration.
Frequently asked questions
What is a liability shield for AI companies?
A liability shield is a legal protection, usually created by statute, that limits or blocks lawsuits against a company for harms connected to its product. In the AI context, developers have sought rules that would prevent them from being automatically held responsible for what their models output or for how customers deploy them downstream.
What exactly did Bessent say?
Treasury Secretary Scott Bessent told CNBC's 'Squawk Box' that the Trump administration will not give AI leaders a liability shield. In the same appearance he discussed AI safety concerns and this week's summit between President Donald Trump and Chinese President Xi Jinping. He did not announce any specific new regulation.
Why does this matter for state AI laws?
A federal liability shield would typically come with preemption language overriding conflicting state statutes. Without one, individual states remain free to write and enforce their own AI rules, meaning a developer selling nationwide must comply with the strictest regime it operates in rather than a single federal standard.
How did AI-heavy stocks close before the comments?
At the last trade on Friday, 18 September 2026, the Nasdaq 100 tracker QQQ closed at $721.45, up 0.63% from a prior close of $716.92. The S&P 500 tracker SPY closed at $761.69, down 0.12%, and the Dow 30 tracker DIA closed at $515.88, down 0.48%. Markets were closed at the time of writing.
Does this mean new AI regulation is coming?
Not directly. Declining to grant immunity is different from imposing rules. The stated position leaves existing courts and state legislatures as the main channels for AI accountability, which is a lighter federal footprint on paper but a less predictable legal environment for developers than either immunity or a clear national statute.
What is the connection to the Trump-Xi summit?
AI firms have argued that legal and regulatory constraints on American developers would advantage Chinese competitors. Bessent raised both AI safety and the summit in the same appearance, indicating the administration is treating AI accountability as a domestic policy question rather than something to be softened for competitive reasons.
