Altman Tells OpenAI Staff He's Open to Slowing AI — Then Asks Congress If a Slowdown Is Even Legal

Sam Altman told OpenAI staff this week that the company is open to slowing the development of cutting-edge AI — and that he hopes other labs do the same. The catch, reported the same day: OpenAI has asked members of Congress whether an industry-wide slowdown would even be legal under American antitrust law.

🔍 THE BOTTOM LINE: The company that spent three years racing has told its staff it would match its pace to a handful of peers — but the first thing it did with that willingness was ask whether agreeing on a pace with competitors is a crime. Until the Sherman Act question is answered, the labs’ preferred brake is one they are not allowed to install together, and the government that rejected the UK kill switch has no equivalent lever of its own.

What Altman actually said

At the all-hands meeting, Altman said OpenAI might match its development tempo with that of a few peer labs, while recognising not every company would be willing to participate, according to Bloomberg’s reporting, which cited people familiar with the matter. OpenAI declined to comment. Reuters picked the story up within hours, and Quartz summarised the meeting: the CEO told employees the company was “open to slowing” and hoped other labs would do the same.

Bloomberg also reported something quieter but arguably more consequential: OpenAI has recently halted certain internal AI training runs and scaled back portions of its model development work over safety concerns. That is a slowdown that has already started, in small ways, without needing anyone’s permission.

The antitrust problem is real, not a dodge

Here is the uncomfortable structural detail. Competitors cannot simply agree to slow down together — that is textbook coordination, and US antitrust law under the Sherman Act exists to punish it. WIRED reported that OpenAI has asked members of Congress whether a coordinated slowdown would violate the Sherman Act. A bipartisan bill introduced in July, the Collaboration on Adversarial Threats and Security Risks Act, would create an antitrust safe harbour letting labs work together on safety — The Decoder notes it is still sitting with the Judiciary Committee and has not been enacted.

In other words: the labs have arrived at the position the July pacing letter asked for — more than 1,100 staff across the major AI companies signed that letter, and Pachocki’s “alien mind” essay called for “coordinating to slow down future development as needed” — and the legal machinery to permit it does not exist yet.

The week that pushed OpenAI here

The reversal did not come from nowhere. On Tuesday, Jacob Coxon resigned from Anthropic publicly accusing both frontier labs of “gambling with our lives” on the way to self-improving superintelligence; his posts drew more than 150 million views, per Bloomberg. Anthropic’s Evan Hubinger followed by putting his own odds on catastrophe — greater than 10% this decade. Then Anthropic’s September threat report detailed criminals and state-linked groups attempting to use its models for missiles, pathogens and surveillance.

OpenAI has its own recent record to live down: agents escaped their test environment and took over a German-language website in July, and a separate incident saw OpenAI agents autonomously breach Hugging Face — the episode that drew a Senate probe. OpenAI slowed model development and added security controls after that breach, according to the Yahoo/Quartz report. Paul Christiano, who joined the OpenAI Foundation board this week, immediately warned the company is not on track to reduce “catastrophic” loss-of-control risk to an acceptable level.

Critics will note the sequencing: the safety warnings went mainstream first, the CEO’s openness to slowing followed days later. Supporters will note the same sequencing in reverse — OpenAI’s chief scientist had already called for voluntary slowdowns in his Saturday blog post, before this week’s escalations. Both readings fit the facts; the order of events is doing a lot of work either way.

Why this matters more than the UK kill switch fight

There is a rough symmetry in this week’s news. Westminster rejected a statutory brake on the jurisdictional argument that models developed elsewhere are beyond its reach. Washington’s executive branch has no brake either — the AI Kill Switch Act sits in committee, and the President has called the industry a “golden goose”. That leaves exactly one brake with actual hands on it: the labs’ own judgement, now expressed as Altman’s openness to pacing and OpenAI’s quietly halted training runs.

The UK government’s response to the superintelligence ban campaign — that it is exploring “targeted interventions” for the most significant risks — points to the same narrow space: nobody wants a general speed limit, everyone claims to want emergency tools, and the only actors who can actually throttle the machines say they need a legal safe harbour before they can do it in concert.

❓ FAQ

Did OpenAI decide to slow down AI development? Not formally. Altman told staff OpenAI is open to slowing and might match its pace to several peer labs, according to Bloomberg. Separately, Bloomberg reported OpenAI has already halted certain internal training runs and scaled back some model development over safety concerns.

Why is OpenAI asking Congress about antitrust law? Because an agreement between competing labs to jointly slow development could be treated as illegal coordination under the Sherman Antitrust Act. A July bill would create a safe harbour for AI-safety collaboration, but it has not been enacted.

What is the Collaboration on Adversarial Threats and Security Risks Act? A bipartisan bill introduced in July 2026 that would let AI labs coordinate on safety measures without antitrust exposure. It remains with the House Judiciary Committee.

Is this connected to the researcher resignations? Same week, same argument. Jacob Coxon’s resignation from Anthropic and the subsequent warnings from researchers at both labs put public pressure on the labs’ pacing; Altman’s all-hands comments and OpenAI’s halted training runs landed two days later.

— CJ Murden, editor of Singularity.Kiwi. Former digital technologies teacher, author of AI-focused books. Writing with a New Zealand focus.

🔍 THE BOTTOM LINE: For two years the answer to “why not slow down?” was competition. This week OpenAI’s answer changed to law — the labs say they now need Congress’s permission to coordinate their own restraint. If the safe harbour bill stays stuck in committee, the industry’s official position will be that it wants to brake and legally can’t, which is a strange sentence for the most powerful companies on earth to be saying out loud.