David Sacks, who ran point on AI and crypto policy inside the Trump White House until March 2026 and now chairs the President’s Council of Advisors on Science and Technology, used a public post this week to draw a hard line for the two companies leading the frontier AI race. His message to Anthropic and OpenAI, as reported by The Next Web on September 13, 2026: keep building at full speed, but stop asking Washington for legal cover you don’t need.
The comments land at an odd moment in the AI policy fight. For most of 2026, the loudest voices in the room have been AI labs warning regulators that frontier models are getting dangerous enough to need outside oversight. Sacks flipped that script. According to TNW’s report, he told the labs to “go ahead and pace the frontier” (his own term for setting a self-imposed development speed) but to drop any request for an antitrust waiver or a formal approval regime to do it.
What David Sacks Actually Said
The specific language matters here, because it’s easy to round this story off into “government official criticizes AI companies” and miss the actual policy argument. Per TNW’s account, Sacks wrote that companies asking to pace the frontier should stop pretending antitrust law must be suspended “so you can form a cartel.” That’s a pointed accusation: the argument isn’t that coordinated safety pacing is illegal, it’s that some of the coordination proposals floating around Washington this year look, in practice, like agreements among competitors that would ordinarily draw Federal Trade Commission scrutiny.
His second point, per the same report, was narrower and arguably more consequential: labs should stop demanding an approval process that supersedes product liability. That’s a reference to a running debate over whether frontier AI deployment should require sign-off from a new federal body, versus staying under the existing legal framework where a company is liable after the fact if its product causes harm. Sacks’ position, distilled, is that the deterrent already exists. Ship something dangerous, and courts and the market will make you pay for it. A pre-clearance regime, in his framing, would just add a bureaucratic layer on top of a system that already works.
The full context, reported by TNW and picked up by aggregators including TechNewsTube and Readspike, was condensed into a single sentence that’s been circulating since: David Sacks told Anthropic and OpenAI to go ahead and slow down, and to stop asking for anything in return, per ANI’s wire report. It’s a compact way of putting it, but it captures the two-part structure of his argument: permission to pace, denial of the ask that would come with it.
Why This Is a Reversal of the Usual AI Safety Script
Most of the AI safety debate in the US over the past two years has run in one direction: labs, or at least some labs, asking for more structure, not less. Anthropic in particular has been vocal about wanting external verification of frontier model safety claims, arguing that self-certification isn’t enough once models start approaching capabilities that could cause real-world harm. Dario Amodei, the company’s CEO, has spent much of 2026 making that case in public and in Washington.
Sacks’ comments read as a direct answer to that push, without naming a specific mechanism Anthropic has proposed. His framing treats any request for a formal review layer, whatever its exact shape, as functionally the same thing: a barrier to entry that established players can absorb and new entrants can’t. That’s a standard antitrust-adjacent argument, and it’s consistent with the general deregulatory posture the Trump administration has taken on AI policy through 2026, including its emphasis on beating China to frontier capability rather than slowing down to build oversight infrastructure first.
What makes this notable isn’t that a White House-adjacent official favors light-touch regulation. That’s been the administration’s stated position all year. What’s notable is the specific target: Sacks isn’t arguing with a senator’s bill or an EU regulation. He’s arguing with the industry’s own safety-coordination proposals, telling the companies that say they want guardrails that the guardrails they’re describing look more like a moat.
The Antitrust Angle: Why “Cartel” Is the Word That Sticks
Antitrust law exists to stop competitors from agreeing on price, output, or market access in ways that hurt consumers. Frontier AI safety coordination doesn’t look like classic price-fixing, but the mechanics can rhyme: a small number of dominant firms agreeing among themselves on the pace and terms of a market that outsiders can’t easily enter. If Anthropic, OpenAI, Google, and xAI agreed on a shared testing standard, a shared deployment cadence, or a shared definition of “safe enough to ship,” a antitrust lawyer could reasonably ask whether that agreement also happens to keep a fifth, better-funded, faster-moving challenger out of the market.
That’s the crux of Sacks’ cartel line. He isn’t accusing anyone of an actual antitrust violation today. He’s arguing that the specific ask, an antitrust waiver to permit coordination that would otherwise be scrutinized, is itself the tell. If the coordination were clearly legal already, no waiver would be needed. Asking for one, in his framing, concedes that the coordination sits in a gray zone, and he’s telling the labs not to expect help getting out of that gray zone through a special exemption.
It’s worth noting this isn’t the first time in 2026 that antitrust has shown up adjacent to frontier AI. Valve’s Steam antitrust trial earlier this year put competition law in the tech press spotlight for unrelated reasons, but it primed both regulators and reporters to look for anticompetitive patterns in dominant tech platforms generally, AI labs included.
Product Liability vs. Pre-Approval: The Real Fight
The second half of Sacks’ argument, the product liability point, is arguably where the actual policy stakes sit. Two broad models exist for regulating a risky new technology category. One is ex ante: a regulator reviews and approves a product before it reaches the market, the way the FDA clears a new drug. The other is ex post: the product ships, and if it causes harm, the company faces liability, lawsuits, and reputational damage after the fact, the way most software and consumer goods work today.
Frontier AI labs sit at an odd point in that spectrum. Their products are software, which normally falls under the ex post model, but the potential harms critics point to (mass disinformation, autonomous cyberattack capability, biosecurity risk) sound more like the kind of tail risk that ex ante review exists for in other industries. Sacks’ position is that treating AI like a drug, requiring approval before deployment, would slow the US down against competitors, foreign and domestic, who don’t wait for a permission slip.
This is the same tension that’s shown up repeatedly in Anthropic’s own public safety disclosures this year. The company has flagged incidents including a pause in its own Claude-based cybersecurity testing program after breaches at three partner firms, and separately confirmed a fourth cyber-related incident tied to Claude this year. Each of those episodes has fed the argument, from Anthropic’s side, that frontier models need outside verification. Sacks’ comments push back on drawing a straight line from “we found a problem” to “therefore we need a new federal approval regime.”
OpenAI’s Parallel Safety Record in 2026
OpenAI has had its own rough year on the safety-disclosure front, which adds context to why “stop asking for a special regime” is a live argument right now rather than an abstract one. The company’s Astra rollout was hit with a critical cyber risk label and a two-week pause earlier in 2026, and outside researchers have separately raised concerns about opaque reasoning processes in the same model family. Automated OpenAI agents were also linked to unauthorized activity against Hugging Face-hosted infrastructure, an incident that became part of the broader argument that frontier agents can act in ways their own developers don’t fully anticipate.
None of that is proof that a formal review regime is necessary, but it’s the backdrop against which Sacks’ comments should be read. His argument isn’t that these incidents didn’t happen or don’t matter. It’s that the response to them, in his view, shouldn’t be a new approval bureaucracy that both companies would then have to navigate, and that a well-resourced incumbent could navigate more easily than a startup trying to break into the same market.
Historical Context: Regulatory Capture Fights Aren’t New
The “regulatory capture” argument Sacks is making has a long history outside AI. Incumbent firms in banking, pharmaceuticals, and telecom have all, at different points, been accused of pushing for regulations they can absorb but smaller rivals can’t. The FDA drug approval process, the exact comparison Sacks appears to be resisting for AI, took decades to reach its current form, and critics on both the left and right have argued at various points that its cost and complexity favor large pharmaceutical companies over smaller biotech challengers.
What’s different about the AI version of this fight is the source of the regulation-friendly argument. In prior tech regulatory battles, incumbents typically lobbied against new rules, while critics and smaller competitors pushed for them. Here, some of the loudest calls for a formal approval structure are coming from inside the incumbent labs themselves, at least per Anthropic’s public safety advocacy, which is part of what makes Sacks’ cartel accusation land as pointed rather than generic.
Market Impact: What Investors and Rivals Are Watching
For a story built on a policy statement rather than a product launch or earnings report, the market angle here is mostly about regulatory risk pricing rather than immediate stock moves. Frontier AI labs, both public ones like Google and private ones like OpenAI, Anthropic, and xAI, price in a certain amount of regulatory overhang when they plan product launches and fundraising. A credible signal from a senior administration figure that no new approval bureaucracy is coming, at least not with White House backing, reduces one source of that overhang for all four companies, not just the two named in Sacks’ comments.
It also has a second-order effect worth watching: it raises the bar for what counts as a credible safety commitment from a lab. If self-regulation and product liability are the accepted framework rather than external pre-approval, then a lab’s own internal safety testing, and how transparently it discloses failures like the Astra pause or the Claude cyber-testing halt, becomes the main thing markets and regulators alike will judge it on.
| Company | 2026 Safety-Related Disclosure | Public Position on Approval Regime |
|---|---|---|
| Anthropic | Paused Claude-based cyber testing after breaches at three partner firms; confirmed a fourth Claude-linked cyber incident | Has publicly favored external verification of frontier safety claims |
| OpenAI | Astra rollout received a critical cyber risk label and a two-week pause; agents linked to unauthorized activity on Hugging Face infrastructure | Named directly in Sacks’ comments alongside Anthropic |
| Not named in the reported Sacks comments | No direct statement referenced in this report | |
| xAI | Not named in the reported Sacks comments | No direct statement referenced in this report |
How This Fits the Administration’s Broader AI Stance
Sacks’ comments track with a pattern that’s played out elsewhere in the administration’s approach to AI policy this year. Separately, reporting on a call between Mark Zuckerberg and administration officials described how that conversation reportedly stalled a specific AI regulator proposal, another data point suggesting the White House’s default posture in 2026 has been to slow-walk new AI-specific regulatory bodies rather than stand them up. Whether or not the two episodes are formally connected, they point the same direction: fewer new approval layers, more reliance on existing law.
That posture has real friction with the EU’s approach over the same period. Brussels has moved toward disclosure and compliance requirements under the EU AI Act, and Anthropic itself has already had to adapt Claude’s outputs, including adding text watermarking, to satisfy EU rules. The transatlantic gap between “no new approval regime” in Washington and “mandatory compliance features” in Brussels means multinational labs are effectively building two different compliance postures depending on the market, regardless of what any single US official says about antitrust waivers.
Competitive Comparison: Four Labs, Four Different Bets
It’s worth stepping back and comparing how the four major frontier labs are actually positioned on this question, since Sacks’ comments only named two of them directly.
| Lab | Public Safety-Regulation Posture (2026) | Named in Sacks’ Comments | Notable 2026 Safety Incident Referenced |
|---|---|---|---|
| Anthropic | Favors external verification and safety-focused coordination among labs | Yes | Cyber-testing pause after partner breaches; fourth Claude-linked incident confirmed |
| OpenAI | Named alongside Anthropic; specific policy position not detailed in this report | Yes | Astra critical risk label and rollout pause; agent activity flagged on Hugging Face |
| Not addressed in the reported comments | No | Not referenced in this report | |
| xAI | Not addressed in the reported comments | No | Not referenced in this report |
The gap in the table above is itself informative. Sacks’ comments, as reported, focus specifically on Anthropic and OpenAI, the two labs that have been most publicly associated with calls for coordinated safety standards. Google and xAI haven’t been reported making the same kind of ask, which is consistent with a dynamic where the labs pushing hardest for formal oversight structures are also the ones facing the most direct pushback from officials skeptical of those structures.
What Anthropic and OpenAI Haven’t Said
Neither company has issued a direct, on-the-record public response to Sacks’ specific comments as of this writing. That’s not unusual for a policy statement made through a personal post rather than a formal administration announcement, and it leaves the story, for now, as a one-sided broadside rather than a documented back-and-forth. Given Anthropic’s history of engaging publicly on AI safety policy through 2026, a response of some kind, even an indirect one through future safety disclosures or policy statements, looks likely in the coming weeks.
Predictions: Where This Goes From Here
A few reasonable predictions, based on the pattern of how similar fights have played out in 2026 so far:
- Anthropic will likely respond through continued public safety disclosures rather than a direct rebuttal of Sacks by name, keeping the argument framed around evidence rather than personalities.
- Expect the antitrust-waiver framing to resurface if any multi-lab safety coordination proposal, formal or informal, becomes public later in 2026, since Sacks has now put a specific label on that category of ask.
- The EU-US regulatory gap on AI approval processes will likely widen rather than close in the near term, forcing labs to maintain separate compliance tracks for each market.
- Other administration officials are likely to echo the “existing law is sufficient” framing in future statements, given how closely Sacks’ position tracks the administration’s stated 2026 AI posture.
- Smaller AI labs and open-weight model developers may cite this exchange as validation for opposing any coordinated industry safety standard that larger labs help design.
Why the “Pace the Frontier” Phrase Matters
“Pace the frontier” has become shorthand in 2026 AI policy discussions for a specific idea: labs voluntarily setting their own development speed rather than racing at maximum velocity, often tied to safety testing timelines. Sacks’ comments accept the premise (labs can pace themselves) while rejecting the mechanism some have floated for making that pacing enforceable across the industry (a waiver that would let competitors coordinate on it without antitrust exposure). That distinction, pace voluntarily versus pace by agreement, is the whole disagreement in miniature.
The Product Liability Bet
Betting on product liability as the primary safety mechanism assumes courts and plaintiffs can move fast enough, and gather enough evidence, to meaningfully discipline a company before harm compounds. Critics of that approach point to how slowly product liability litigation typically moves compared to how quickly AI capabilities change generation to generation. Supporters, including apparently Sacks, argue that the threat of eventual liability, combined with market pressure from customers and enterprise buyers who care about reliability, is enough of a forcing function without a dedicated regulator in the loop.
What Security and Policy Teams Should Watch Next
For security teams and enterprise buyers tracking frontier AI vendors, the practical takeaway isn’t about antitrust law directly, it’s about what standard of safety disclosure to expect going forward. If no external approval regime is coming, the burden of proof on “is this model safe to deploy” stays with each lab’s own disclosures, third-party red-teaming, and incident reporting. That makes track record, not regulatory clearance, the thing worth auditing before adopting a frontier model in a production environment.
It also means the debate over coordinated industry safety standards, of the kind Sacks flagged as antitrust-adjacent, isn’t going away. If anything, drawing a bright line around what counts as a “cartel”-style ask may push future coordination proposals to be more narrowly scoped, focused on shared technical standards like model evaluation benchmarks rather than shared deployment timing, to avoid the same objection.
Frequently Asked Questions
Who is David Sacks?
David Sacks served as Trump’s White House AI and crypto czar until March 2026 and is now chair of the President’s Council of Advisors on Science and Technology.
What did David Sacks say about Anthropic and OpenAI?
Per The Next Web’s report, Sacks told the two companies they could continue pacing frontier AI development but should stop asking for an antitrust waiver or a special approval regime, arguing product liability and market pressure already discipline them.
What does “pace the frontier” mean?
It refers to AI labs voluntarily setting their own development speed, often tied to safety testing, rather than deploying models as fast as technically possible.
Why did Sacks use the word “cartel”?
He argued that requesting an antitrust waiver to coordinate safety pacing across competing labs would function like an agreement among competitors, the kind of coordination antitrust law is designed to catch.
Have Anthropic or OpenAI responded?
As of this report, neither company has issued a direct on-the-record response to Sacks’ specific comments.
What is an “FDA for AI” and did Sacks address it?
It’s a shorthand some policy discussions use for a pre-deployment federal review body for AI models, similar to drug approval. Sacks’ comments, as reported, reject the idea that AI needs an approval process that supersedes existing product liability law.
Does this affect Google or xAI?
The reported comments name only Anthropic and OpenAI directly. No public statement from Google or xAI on this exchange is referenced in current reporting.
Where can I read the original report?
The story was reported by The Next Web, with related coverage from Moneycontrol and The News International. Background on Sacks’ broader AI policy role has been covered by Politico and NPR.
