Bessent denies the labs a liability waiver and pushes open models

Share
Bessent denies the labs a liability waiver and pushes open models

Two AI-policy answers landed in the same Washington hearing: what the frontier labs cannot have, and what the Treasury Secretary wants built instead.

Treasury Secretary Scott Bessent told the House Financial Services Committee on Tuesday that the government should not give AI companies a liability exemption, and that the United States needs to build more open models of its own. "I can tell you what we shouldn't do on safety is give these labs a liability exemption, which is what they're asking for," Bessent said, arguing that "the best way to guarantee safety is that the creators are liable for what they build and generate," and that the industry's position reduces to "we would like to all slow down, but please give us a waiver on liability." He told both chambers not to consider one: "The one thing we should not do is give them a blank check on liability." His second answer was industrial rather than regulatory — "we need to develop more open source models in the U.S. We can't let these large labs have regulatory capture because that will stop innovation" — which puts the Treasury on the same side as the open-weights camp in a fight the labs have been framing as a safety question.

Bessent also carried an allegation into the hearing: "Kimi had a breach and sent Chinese weapons plans back to Anthropic," he said, a reference to Anthropic's September threat-intelligence report, which alleged that Moonshot routed some Kimi users' queries through Claude. Anthropic did not immediately respond to a request for comment. On the distillation argument, his version was blunt — "the technical word for stealing and copying American AI models is distillation," and a lab that copies homework "can never get a higher grade" — with the policy conclusion that American open models would eventually crowd Chinese ones out of the market.

The tell is which item of the labs' wish list Bessent refused. A slowdown costs a lab revenue it can measure, but a liability exemption is the one ask that transfers risk to everyone else, and refusing it is free for a politician. It also matches where the rest of the administration has landed in the past week: the FTC chair called the safety-for-antitrust trade a request for "barriers to entry," and the President spent Monday telling a live audience that safety fears are a hoax. The labs are asking two branches of government for two different things — cover for coordinating, and a federal framework that binds their rivals — and both answers so far have been no. We tracked the antitrust half of that ask — FTC chair is 'deeply suspicious' of the labs' antitrust ask.


More than 100 people rallied outside Philadelphia's City Hall on Monday to launch a campaign against data centers, and the argument they are making is about pollution, bills and noise rather than about models. The No Data Centers in Philly campaign was organized in a city still living with the aftermath of its refinery — Philadelphia Energy Solutions, its largest, closed in 2019 after an explosion — and Philly Thrive's Sonya Sanders made the connection explicit: "We don't want the pollution, we don't want the noise, and we know they're not going to bring us more jobs." Two sites are already in play in the city. In New York, Governor Kathy Hochul put the same test in policy language at a press conference: "Progress shouldn't arrive with a higher utility bill, deleted water supply, or noise pollution. These data centers can only be built, should only be built in places that want them."

The pattern is repeatable because the objections are local and identical everywhere: Denver, Indianapolis, Asheville, Charlotte and Reno have all put moratoria on new builds in front of councils this year, and a Gallup poll this year found a majority of Americans opposed to data centers near where they live. TechCrunch's framing is the sharper one — opposition burns hottest in towns that already paid for heavy industry once. For the industry, that is a worse problem than a permit fight, because the opposition is not against AI, it is against being asked twice. We looked at the demand side of the same argument — AI data centers will out-burn Germany and Japan on gas by 2035.

What to watch: whether the FRONTIER Act's preemption clause and Treasury's open-model push end up in the same bill, and whether any of the labs answers Bessent's liability refusal on the record.

If liability is the one safety lever that costs the labs real money, why is refusing it the easiest answer in Washington? Tell us in the comments.

Read more

Houmo moves compute inside memory — and A20 enters production

Houmo moves compute inside memory — and A20 enters production

Beijing chipmaker Houmo used its partner conference this week to lay out two compute-in-memory routes at once — and to say the first chip built on the newer one is already in production. Houmo's A20, its first chip built on a 3D compute-in-memory architecture, has entered production. The part supports up to 16 chips fully interconnected, claims a 50% improvement in compute utilization over the previous generation, and covers models from 3B to 30B parameters, according to co-founder and product

Kling 4.0 goes live with 30-second native video at 10-bit HDR

Kling 4.0 goes live with 30-second native video at 10-bit HDR

Kuaishou's Kling AI has spent two years becoming the video model people actually finish projects with. Its new flagship, Kling 4.0, went fully live on October 15 — and this round is less about prettier frames than about the two things that make a generated shot usable in a real edit: length and control. Kling 4.0 generates up to 30 seconds in a single native pass, outputs 10-bit HDR at 4K, 1080p and 21:9, and takes up to 15 reference assets in one task. That budget splits across ten images, fi

Brain organoids classified Braille at 83% — in a trio, not alone

Brain organoids classified Braille at 83% — in a trio, not alone

Two stories today: human brain tissue on a chip decoded a tactile signal, and the data-centre company behind some of the largest AI contracts in Europe turns out to be six months behind on its flagship site. Human forebrain organoids grown on a microelectrode array classified Braille letters with 61 percent accuracy on their own and 83 percent when the responses of three organoids were combined, in a peer-reviewed study from the University of Bristol. The team took event-based tactile readings

Suleyman says Microsoft's base model beat OpenAI's on a quarter compute

Suleyman says Microsoft's base model beat OpenAI's on a quarter compute

Microsoft finished pre-training its biggest in-house model yet — then used it to make a point about the company it bankrolls. Microsoft AI completed pre-training of MAI-Base-2, a mixture-of-experts foundation model with 80 billion active and 2.8 trillion total parameters, and says the recipe is 4.2–12.3x more pre-training FLOPs-efficient than DeepSeek-V4-Pro-Base. The run itself is the number worth keeping: 16,000 Nvidia GB200 GPUs for 38 days across August and September, at 21% average model