Invest1 publisher2 min readPublished
Bessent leaves the cost of AI system failures with the developers
Treasury Secretary Scott Bessent said on October 3 that the government will not give AI developers a liability shield. Anyone holding those developers' shares or debt now has to price the cost of a failing system as the company's own.
The Investor · Invest desk
What happened
- Bessent called AI leaders' existential-risk warnings alarmist, arguing that sounding the alarm without workable answers does not amount to leadership.
- He urged AI companies to take proactive steps on their own instead of waiting for lawmakers to impose restrictions.
- The refusal matches warnings against liability waivers that Bessent set out in a series of statements in September 2026.
- Bessent has raised AI policy concerns since spring 2026, after warnings about risks to critical infrastructure.
- The administration says it wants to guard against advanced models that could undermine the financial system while keeping the US technological lead.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- exposure The developer keeps the legal and financial exposure when a system fails, so its shareholders and lenders are the parties left absorbing any damages.
- contradiction Bessent's September call for a voluntary pause on frontier models pulls against the goal of staying ahead of China, and Crypto Briefing cannot say how the two fit.
- precedent Crypto Briefing expects Treasury to keep examining how advanced models interact with finance, and that scrutiny reaches banks and market participants as well as the labs.
In Crypto Briefing's account, Bessent compared the control-risk debate to Hannibal Lecter, and the line it quotes is "Stop me before I kill again." [3] That line asks someone else to do the restraining. Bessent's answer on liability is that accountability rests with humans and not with AI systems [6]. A shield would have moved part of the cost of a harmful output off the company that shipped the model. The official who oversees the US financial system [16] has said Washington will not grant one [1].
So Treasury is turning down two jobs. Its preference for self-policing means fewer mandated rules, and Crypto Briefing says that can read as friendly to innovation [13]. Its refusal of a shield means it will not carry the losses either. One reading follows from that pairing. If self-regulation spares the labs restrictions that would have cost more than the legal exposure they keep, the policy works out in their favour. The interview, as reported, did not put a figure on either half.
A second reading is that nothing has moved. Crypto Briefing did not report that a shield was ever on offer, so a developer's exposure after the interview is the same as it was the day before. I think that is right about the present and wrong about the forecast. Bessent's stance against waivers has now held for two consecutive months [1]. Anyone valuing an AI developer over a decade should give a federal backstop a lower probability after two refusals than after one, and should put the legal exposure into the developer's cost of capital.
The third reading runs through Washington's talks with Beijing on AI safety mechanisms [11]. If those talks, or any other part of the government, produce a safe harbour for developers, the view here is wrong.
What to watch
- Disclosure by AI developers of litigation reserves or liability insurance costs. Those figures would put a number on the exposure Treasury left with them.
- Whether any frontier lab announces a pause on model releases in response to Bessent's September call.