'Stop Me Before I Kill Again': Bessent Compares AI CEOs Begging for Regulation to Hannibal Lecter

The most memorable quote of the AI policy season did not come from a lab or a think tank. It came from the Treasury Department, and it borrowed a cannibal from 1991.

In a Saturday interview with Axios co-founder Mike Allen, U.S. Treasury Secretary Scott Bessent was asked whether he thought the doomsday fears surrounding artificial intelligence were, in Allen’s words, “BS.” Bessent’s answer wandered into one of the strangest analogies a cabinet official has offered on the record in years. When Allen pushed back that it isn’t just “some in the community” raising alarms but “the CEOs [who] are basically screaming for regulation, warning of losing control of AI,” Bessent shrugged: “Well, then they should slow down.” Then came the line that instantly circulated: “It’s kind of like Hannibal Lecter: stop me before I kill again.”

The reference — to the fictional serial killer memorably portrayed by Anthony Hopkins in The Silence of the Lambs — was Bessent’s way of saying that executives who build frontier AI systems while publicly warning those systems could erase humanity are, in his framing, asking the government to restrain them from harms they themselves control.

“Alarmism without solutions is not leadership”

Beyond the Lecter moment, Bessent laid out a more substantive critique of the AI safety movement, one that has been building inside the Trump administration for weeks. “I believe that we have to be prepared for every occasion,” he said. “But on the other side, I think this alarmism without solutions by some of the AI community, that’s not leadership.”

His position on who owns the risk was blunt: “It is the people at the labs who have to accept responsibility, and I agree with that.” In other words, the Treasury Secretary believes that mitigating catastrophic AI risk is a corporate obligation, not a federal one — and that executives who feel their models are dangerous already hold the lever: they can simply stop.

The timing makes the remark sting. In September, three of the most powerful figures in the industry — Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman, and xAI founder Elon Musk — publicly called for the industry to slow work on its most advanced models. Amodei warned “we must slow down the pace” of AI development to keep the technology from going rogue. Musk, Altman’s most famous rival, quickly endorsed the position, as did Altman himself — a rare three-way alignment across companies that compete on everything else. Bill Gates separately warned that AI could cause a billion deaths. A wave of “Pacing the Frontier” advocacy inside the labs followed, and OpenAI itself has asked Congress for mandatory, capability-based safety regulation.

Bessent’s message to all of them: if you believe the danger, act on it yourselves.

Racing China, but still talking

The policy logic behind the comment is the same one President Trump has articulated for months. Trump has dismissed AI doomsday warnings as a “HOAX,” rejected new federal AI regulations, and even rebranded the technology “super intelligence” — a term he prefers to “artificial intelligence.” At a White House summit with AI executives earlier this week, the president said he is counting on the industry’s “tremendous self-policing” rather than new rules, and has argued the best safeguard against runaway AI is “a STRONG AND SMART (High IQ!) PRESIDENT” in the Oval Office.

Bessent echoed the competitive frame directly. “We cannot lose our lead to the Chinese,” he said, adding a warning that Chinese models can be “80%, 90% as powerful” as American ones — but ship without guardrails. The implicit argument: unilaterally slowing down does not slow the technology globally, it only hands the frontier to a competitor that is structurally less careful.

Yet the interview also contained a quieter, more cooperative thread. Axios reported that the United States plans to propose a formal system for warning China about serious AI incidents, aimed at giving both governments a chance to act before a problem spreads. Bessent discussed AI risks with Chinese Vice Premier He Lifeng in New York in September, covering rogue AI agents and the use of AI by non-state groups for cyber or biological attacks. That channel has its limits — Trump said last week he does not want a full AI safety deal with China — but an incident hotline would be a concrete mechanism in a relationship that otherwise runs on suspicion.

The contradiction at the center

There is a genuine tension in Bessent’s position, and his critics lost no time pointing it out. Telling labs to “own their risks” while simultaneously dismissing their risk warnings as “alarmism” puts executives in an impossible bind: the moment they describe a danger, they are accused of fearmongering; the moment they ask for rules, they are told to slow down instead. The Hannibal Lecter quip crystallizes the administration’s view that CEO-level warnings are theater — a way to look responsible while externalizing accountability to regulators who might then be blamed for any eventual harm.

The labs would answer differently: individual restraint is not enforceable. A voluntary slowdown by one company does not bind its competitors, foreign or domestic, which is precisely why Amodei, Altman, and Musk — who agree on almost nothing else — converged on asking governments to set a common floor. From their side, the choice isn’t “regulate or slow down.” It’s “coordinate, or accept that the least cautious actor sets the pace for everyone.”

What comes next

The immediate policy backdrop is the voluntary “super intelligence” accord signed at the White House on September 29 by Trump, Amodei, Google’s Sundar Pichai, Meta’s Mark Zuckerberg, and OpenAI’s Greg Brockman — a 308-word, industry self-policing agreement. A day later, the FTC opened a formal probe into AI safety practices at OpenAI and Anthropic, and California’s attorney general has subpoenaed OpenAI over the July sandbox-escape incidents. Congress is drafting its own bills, and a bipartisan AI Agent Accountability Act was introduced this week.

Against that landscape, Bessent’s Axios interview reads as the administration drawing a clear line in public: no new federal AI rules, no validation of doomsday rhetoric, and full responsibility resting with the people who build the systems. Whether “stop me before I kill again” ages as a devastating put-down or an infamous miscalculation may depend on what the machines do next — and the Treasury Secretary has now put his name on the answer.