AI firms are racing to slow their own development, and antitrust law is doing everything it can to stop them.
The “slowdown” movement has gathered momentum since agents swarmed websites and coordinated via secret message boards. Leading AI companies have called for a coordinated pause in development, but the law treats collusion as a crime. Companies worry that a slowdown would run afoul of antitrust laws, and they are training employees not to use the word “slowdown” at all.
The problem is simple. When companies agree to restrain their own trade, they trigger a federal statute. The Sherman Act, a key US antitrust law, generally bars agreements among competitors to restrict output or raise prices. A slowdown looks like both.
The Sherman Act Problem
John Bergmayer, legal counsel for the nonprofit Public Knowledge, says companies boxed themselves in with their own phrasing. Regulators usually look at whether output is reduced and whether companies are taking it easy.
That is the core of the mess. An AI slowdown reduces output by design. It is also a collective agreement among competitors to act together. Both trigger the Sherman Act.
There is an escape hatch, but it requires permission from the government. An official all-clear could ward off costly investigations. Without it, companies could face scrutiny if regulators see the slowdown as a collusive agreement to reduce trade.
Zuckerberg’s Narrow Defense
Meta CEO Mark Zuckerberg chimed in without endorsing an explicit slowdown. He argued that AI labs have a natural incentive to make agents behave better to avoid misalignment.
That is a defense, but it is a narrow one. It assumes the market will fix safety problems on its own, which is not the same as agreeing to slow down.
The Ancillary Restraints Doctrine
David Lawrence, former policy director of the DOJ’s Antitrust Division, wrote that agreements to prevent catastrophic risks increase output and promote competition under the ancillary restraints doctrine. That is a legal argument, not a policy position, and it cuts against the usual antitrust approach.
The doctrine protects some restraints if they are necessary to achieve a legitimate business purpose. A safety-focused slowdown might fit that logic, but only if the government agrees.
Quality Fixing and the Car Case
Roger Alford, Notre Dame Law School professor and former DOJ deputy chief, warns of a different danger. Collective agreement not to implement safety measures could expose labs to “quality fixing” charges.
Alford cites a European antitrust case where car companies worked together to develop emissions-reducing technology but agreed to not compete on improvements beyond what the law required. They ultimately had to pay roughly the equivalent of a billion-dollar fine.
The Duopoly Accusation
David Sacks, cochair of the President’s Council of Advisors on Science & Technology, accused Anthropic and OpenAI of being a duopoly and called the antitrust exemption request an “election-season psyop.” His accusation remains unresolved.
Sacks is not arguing against safety. He is arguing against the framing. The exemption request, he says, is political theater disguised as a technical concern.
IPO Timing
Anthropic filed confidential IPO paperwork and is expected to go public next month. OpenAI CEO Sam Altman delayed the IPO until 2027 due to safety concerns. OpenAI’s CFO previously told employees the company would be public in 2027, per CNBC.
These dates matter because they put the slowdown in context. Companies are moving toward public markets at a time when safety concerns are growing.
How Companies Are Preparing
Google trained employees not to use phrases like “slowdown” or “pause” to avoid implying anticompetitive behavior. That is a defensive move, not a denial of the idea. The company is preparing for the possibility that the slowdown will be treated as an agreement, so it is scrubbing the language.
What Happens Next
The slowdown movement has not died. It has simply become more complicated.
- Companies will continue to push for a coordinated pause.
- Regulators will watch closely for any signs of collusion.
- The government will need to decide whether a safety-focused slowdown fits the ancillary restraints doctrine.
- Companies will train their employees to speak carefully around the topic.
This is not a fight between good and evil. It is a fight between two legitimate interests: safety and competition.
Where the paper stands
The paper backs AI companies that want to keep developing, and is against both the slowdown movement and the antitrust law that could punish them for agreeing to it. The Sherman Act bars agreements among competitors to restrict output or raise prices, and a coordinated pause fits both descriptions. But the paper’s position is that regulation should target direct harm, not broad restraint, and that the biggest firms should not get to write the rules that protect them from competition.
The story shows the tension clearly. Companies are training employees to avoid saying “slowdown” to stay clear of antitrust trouble, even as they push for a coordinated pause. That is a defensive move, not a denial of the idea. The ancillary restraints doctrine offers a possible escape, but it requires government permission. The paper’s broader point is that narrow rules against direct harm — forcing companies to disclose safety failures they hid — are the right approach, not broad rules that hand the market to the incumbents.
The story points to a deeper problem than just the slowdown. Big tech firms are asking for exemptions from antitrust law, and the paper’s position is wary of any arrangement that only the biggest firms could afford. When the biggest firms ask to be regulated, the paper asks who those rules would lock out. Licensing regimes and compliance costs are a moat, not a safeguard. The reader should watch for who gets to define what counts as restraint, and who pays for the privilege of staying within it.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.