OpenAI's Antitrust Panic Over an AI Slowdown

The Question Nobody Expected

OpenAI has done something unusual for a company at the center of the AI arms race. It has gone to Congress and asked whether coordinating with its competitors to slow down frontier AI development would violate antitrust law.

It sounds almost absurd at first glance. A company that spent years racing to build the most powerful models now wants to know if pulling the brakes collectively is even legal. The very notion of a cartel-like arrangement — only this one aims to restrict output rather than inflate it — should set off every instinct in any competition lawyer’s body. Yet here it is, posed in good faith by the single organization most responsible for accelerating the pace of the race itself.

But that absurdity is precisely the point. The AI safety movement has spent years warning about uncontrolled development. The most prominent voices inside OpenAI — Ilya Sutskever, Noam Brown — have publicly called for pauses. Now they are discovering that the legal infrastructure governing corporate behavior was not designed for this scenario. The Sherman Act, drafted in 1890 to combat the railroad trusts and standard oil monopolies, has no chapter for existential risk mitigation.

The Collision Course

Antitrust law in the United States is built on a simple premise: competitors should not collude to restrict competition. That is why companies cannot meet in a boardroom and agree to slow production, divide markets, or coordinate pricing. Such coordination is per se illegal under Section 1 of the Sherman Act, regardless of whether the outcome seems socially beneficial. The courts have consistently rejected the defense that anticompetitive conduct can be justified by externalities or public welfare concerns. The logic is stark — if you allow one exception, you open the door to countless others, and the enforcement apparatus collapses under the weight of subjective balancing tests.

The OpenAI inquiry reveals a genuine policy trap. If every major AI lab agreed to temporarily halt frontier model training to assess safety risks, the DOJ and FTC would likely view that as a horizontal conspiracy — the textbook definition of antitrust violation. The laws do not distinguish between a cartel fixing prices and a cartel fixing speed. Both are restrictions on output. Both are naked restraints of trade. The Second Circuit’s analysis in cases like American Colloid confirms that agreements to limit production, even for safety reasons, fall squarely within the prohibition.

This is not a hypothetical. People close to OpenAI confirmed to WIRED that the company has raised this directly with lawmakers. The inquiry itself suggests the leadership team is grappling with a contradiction that no one has adequately resolved: the safety case for coordination collides with the legal case against it. Legal scholars who have reviewed the question have been blunt. Ben Cravat, an antitrust lawyer at Jones Day, called the request “kind of insane” and noted that any coordinated pause would be “obviously and undeniably an antitrust violation.” Others have pushed back, suggesting the legal analysis might be more nuanced if framed as information-sharing rather than explicit coordination. But the core tension remains.

Who Wins, Who Loses

The immediate loser is the AI safety movement. If the safest path to coordinating a slowdown is illegal, then the movement’s most dramatic policy proposal — a voluntary industry pause — is effectively neutered by existing enforcement frameworks. Lawsuits would follow. Consent decrees. Criminal referrals. The threat of enforcement creates a chilling effect that makes coordination impossible without government authorization. We have seen this dynamic play out before. Industry groups routinely self-censor on collaboration when antitrust exposure is high, even when the conduct would be harmless or beneficial. The mere possibility of prosecution is enough to deter the very behavior the conduct might protect against.

The winner is the status quo. Companies already committed to rapid development face no legal barrier to continuing. The competition itself remains unconstrained by any collective pause mechanism. In effect, antitrust law becomes the default guardrail — not because it addresses safety concerns, but because it prevents the kind of coordinated response that safety advocates are proposing. The law, in other words, protects the pace of competition while remaining silent on the consequences of that pace.

There is a secondary effect that deserves attention. The inquiry itself — the act of asking Congress whether coordination would be illegal — may carry competitive consequences beyond the immediate legal question. Competitors who learn of the request may interpret it as a signal about OpenAI’s willingness to slow down, which could alter their own strategic calculations. If OpenAI is signaling caution, rivals may interpret that as an opportunity to accelerate and capture market share. The pause movement, if leaked, could paradoxically intensify the very race it seeks to moderate. This dynamic is well-documented in oligopoly theory: signals about restraint are often read as weakness by competitors who benefit from unfettered aggression.

What This Means for Policy

The OpenAI inquiry is a symptom of a deeper problem. Regulators built antitrust enforcement for the industrial economy, where collusion typically harmed consumers through higher prices. AI safety concerns are qualitatively different — they involve existential risk assessment, not price-fixing. Yet there is no statutory exception for coordinating on safety grounds. The Sherman Act makes no distinction between collusion that raises prices and collusion that might, hypothetically, reduce catastrophic risk. The courts have not created one either.

Congress now faces a choice that will define the regulatory landscape for AI. It can either create a narrow safe harbor for safety-related coordination among frontier labs, or it can let antitrust law continue to block whatever collective action safety advocates propose.

The first option requires legislative precision that Washington has struggled to deliver on anything. Crafting a safe harbor that permits safety coordination without becoming a loophole for market allocation would be extraordinarily difficult. Where do you draw the line between legitimate safety consultation and impermissible output restriction? Who defines what counts as a credible safety risk? What oversight mechanism prevents the safe harbor from becoming a permanent cartel arrangement? These are not easy questions, and the legislative history of tech regulation suggests Congress has little appetite for the granular, technical work such a framework would demand.

The second option means the default position remains unlimited competition — with all the risk that entails when the product is increasingly autonomous systems. This is not a theoretical concern. The military applications of AI, the potential for autonomous weapons, the concentration of capabilities in a handful of labs — these are not outcomes that market competition alone is designed to manage. Antitrust law was never meant to be a comprehensive technology governance tool.

There is a third path that has received comparatively little attention: regulatory authorization. Rather than relying on Congress to write a statute, the DOJ or FTC could issue guidance — or even formal interpretive guidance — clarifying that coordinated safety assessments are not enforcement priorities. This would not require new legislation. It would require political courage and a willingness to confront the uncomfortable reality that existing law may be blocking the very coordination it was never designed to address. But agencies have precedent for this kind of flexibility. The Antitrust Division’s business review letters, the FTC’s safe harbor statements — these mechanisms exist precisely for situations where the law’s blunt instrument needs a surgeon’s touch.

What Happens Next

The implications extend beyond OpenAI’s immediate question. If the company proceeds without congressional guidance and attempts to coordinate a pause, it opens itself to enforcement action from rivals who have every incentive to exploit the situation. Competitors benefit from keeping OpenAI’s hands tied. A rival lab that accelerates past a coordinated pause would gain significant first-mover advantage — model capability, talent acquisition, customer lock-in — while the pausing companies absorb the costs of risk reduction without capturing the upside. The economic incentives are misaligned with the safety incentives, and antitrust law reinforces that misalignment.

We should also consider what this moment reveals about the internal dynamics of OpenAI itself. The company has long been structured around a dual mandate — to advance AI research while ensuring that artificial general intelligence benefits all of humanity. Those missions are not always aligned. When the board removed Sam Altman in November 2023, it was partly driven by concerns about pace and safety. The subsequent reversal and Altman’s return signaled that the commercial and strategic pressures had won out. This inquiry may represent a late-stage attempt by the safety-oriented faction to find a legal pathway for the constraints they believe are necessary. Whether they succeed depends less on antitrust doctrine than on political will.

The timing of the inquiry matters as well. Frontier model capabilities are advancing on a curve that exceeds most regulatory forecasting. Each major release increases the concentration of power and the potential for misuse. The window for effective coordination may be narrowing not because of legal barriers alone but because the economic and strategic incentives to accelerate continue to compound. Every month of delay raises the probability that a lab will defect from any informal understanding and seize competitive advantage.

The Bigger Picture

What makes this story significant beyond the immediate legal question is what it reveals about the state of AI governance. The industry is self-aware enough to want to coordinate on safety. The legal system is rigid enough to forbid exactly that coordination. Government has not stepped in with a framework that resolves the tension.

That gap is where the real story lives. The OpenAI inquiry is not just a company asking for clarification — it is a signal that the most powerful AI lab has looked at the regulatory landscape and concluded it needs help from Congress to do what its own leaders believe is the responsible thing. The request is, in a sense, an admission that the existing tools are insufficient. Antitrust law can stop collusion, but it cannot build a safety architecture. It can punish coordination, but it cannot design the alternative.

How Washington responds will determine whether AI safety advocacy operates within the existing legal order or requires a fundamental restructuring of how competition policy treats emergent technological risk. The stakes extend well beyond the AI sector. If Congress refuses to adapt antitrust enforcement to the realities of coordinated safety risk, it sets a precedent that applies to every emerging technology where market competition and collective survival are in tension — bioweapons research, climate engineering, autonomous systems more broadly. The Sherman Act was not written for a world of existential technological risk, but we are living in that world now. The question is whether the law will bend before the circumstances change it.