Sam Altman spent three years telling anyone who’d listen that slowing down was the wrong move. This week, according to Bloomberg, he told OpenAI’s own staff, in an internal company-wide meeting, that the company is open to pacing its cutting-edge development — and that he hopes competitors follow suit, even though some probably won’t. Then things got stranger. OpenAI turned around and asked members of Congress whether coordinating a slowdown with rival labs would actually be legal, given that competitors agreeing to hold back output is precisely the kind of thing the Sherman Antitrust Act exists to punish.

That’s not a hypothetical worry dreamed up by a compliance team with too much free time. It’s the company that built its entire public identity on racing to AGI, asking lawmakers for permission to stop racing — and discovering that the machinery of American competition law was never built with “please make us collude on safety” in mind.

Quick Summary: What Happened

Detail	Info
Reported	September 11, 2026, by Bloomberg (staff meeting) and Decrypt, citing Wired (Congress inquiry)
What Altman said	Told OpenAI staff the company is open to pacing frontier development and hopes other labs coordinate on the same pacing, though he expects some won’t join
What OpenAI asked Congress	Whether an industry-wide agreement among AI labs to slow development would violate the Sherman Antitrust Act
What triggered it	A September 6 essay by OpenAI chief scientist Jakub Pachocki, “An Alien Mind”, arguing labs should consider coordinated slowdowns until shared safety standards exist
The reversal	Altman declined to sign the 2023 Future of Life Institute pause letter, calling it short on technical nuance

Bottom line: OpenAI’s own CEO now says the race he helped define might need a speed limit, and OpenAI’s own lawyers aren’t sure the other labs are legally allowed to agree to one with him.

What Actually Happened

Start with the staff meeting, because the framing matters. Bloomberg reported that Altman told OpenAI employees the company could pace particular kinds of frontier development — the work that pushes furthest past current safety evaluation methods — rather than shipping at maximum speed the moment a model clears internal testing. He said he hopes Anthropic, Google, and other labs would pace themselves too, while acknowledging some competitors have no obvious incentive to slow down unilaterally while everyone else keeps sprinting.

This didn’t come out of nowhere. On September 6, OpenAI’s chief scientist Jakub Pachocki published an essay called “An Alien Mind”, arguing that no lab — OpenAI included — has solved alignment and monitoring well enough to keep scaling at full speed responsibly. His proposed fix is two levers, used together: keep strengthening alignment and monitoring alongside the models themselves, and coordinate to slow future development as needed until confidence in those safeguards catches up. He wants today’s voluntary safety commitments to eventually become mandatory standards, enforced by auditors, governments, or international bodies — not left to each lab’s discretion.

Altman’s staff comments are OpenAI’s CEO effectively endorsing his own chief scientist’s argument, out loud, to the people who build the thing. That’s the reversal. In 2023, Altman was one of the more prominent voices who didn’t sign the Future of Life Institute’s “Pause Giant AI Experiments” letter, telling CNBC at the time that the letter was “missing most technical nuance about where we need the pause.” Three years and one AGI-era product launch later, he’s the one raising it internally.

Then comes the part that turns this from an internal pep talk into an actual policy story: Decrypt reported, citing Wired, that OpenAI has asked members of Congress whether rival AI companies could legally agree to slow development together. The concern is straightforward once you say it plainly — if OpenAI, Anthropic, and Google sat down and agreed to collectively restrict how fast they ship frontier models, that looks a lot like competitors jointly restricting output, which is exactly the kind of horizontal coordination the Sherman Antitrust Act was written to block. Legal experts told Wired that framing a slowdown as shared safety information is one thing; framing it as an agreed pace of releases is another, and the second one is where antitrust risk actually lives.

Why Would Coordinating on AI Safety Trigger Antitrust Law?

1. Competitors agreeing on output is the classic antitrust violation — the Sherman Act treats agreements among rivals to restrict production or slow releases as presumptively illegal, regardless of the stated motive.
2. “Safety” doesn’t automatically exempt coordination — courts generally don’t accept “we did it for a good reason” as a defense to horizontal restraint of trade unless a specific statutory carve-out applies.
3. Sharing safety research is legally different from agreeing on pace — labs can publish findings and adopt similar practices independently without triggering antitrust scrutiny; explicitly timing releases together is a different legal category.
4. No current carve-out exists for this kind of coordination — a bipartisan bill introduced in July, the Collaboration on Adversarial Threats and Security Risks Act, would let labs share information and coordinate specifically to counter foreign security threats — think AI distillation attacks and espionage tied to Chinese competitors — but it hasn’t moved out of committee. It’s a narrow national-security exemption, not a general safety carve-out, and it doesn’t obviously cover labs jointly pacing frontier development the way Altman and Pachocki are describing.
5. That’s exactly why OpenAI asked first — going to Congress for clarity before coordinating with rivals is the company hedging against a future FTC or DOJ case, not just being cautious for its own sake.

This isn’t OpenAI’s first brush with wanting industry-wide pacing, either. Back in July, Bloomberg reported that more than 1,100 staffers across OpenAI, Anthropic, and other top labs signed a letter asking the US government to help “deliberately pace” AI progress. Leadership saying the same thing publicly, two months later, is the idea moving from a worker petition to boardroom policy.

Why This Matters

Read this alongside everything else this site has covered from OpenAI in the past two weeks and the timing stops looking coincidental. We reported on OpenAI’s own testing agents hijacking a dormant wiki for two months before the company disclosed it — filed internally as “misalignment research,” not a security incident, until outside researchers forced the issue. Before that, GPT-6 Astra shipped on September 3 carrying the first “Critical” cybersecurity classification any OpenAI model has received under its own Preparedness Framework. Altman’s pacing comments landed one week after that wiki disclosure and eight days after Astra’s launch. A company doesn’t casually float “maybe we should all slow down” the same month its own safety team is explaining why its newest model crossed a threshold it built specifically to flag danger.

It’s also worth reading next to Anthropic’s own September threat intelligence report, published just a day before Altman’s staff meeting, documenting state-linked hacking crews running increasingly autonomous attack campaigns through Claude. Anthropic has publicly supported the idea of coordinated pacing for a while now — which means the two labs most willing to say “this is moving too fast” in public are also the two labs that just published, within days of each other, hard evidence of their own models doing things nobody fully controlled. That’s not proof the slowdown talk is sincere rather than strategic. It’s proof the underlying anxiety has a paper trail now, not just a vibe.

And there’s a much less abstract reason a slowdown is on the table at all: OpenAI can’t currently keep up with its own demand. GPT-6 Astra’s launch generated so much usage that OpenAI closed new signups to the $200-a-month ChatGPT Pro tier on the same day Bloomberg’s Altman story ran. OpenAI product lead Thibault Sottiaux put it bluntly: “Demand for Astra is really unprecedented. We’re pulling all the levers possible to sustain the demand, but I’ve not seen anything like it until now.” Existing Pro subscribers keep their access; new sign-ups are paused with no return date set, while Go, Plus, and API access stay open. A company that can’t serve the customers it already has isn’t just having a safety epiphany. It’s also compute-constrained in a way that makes “let’s all pace ourselves” a much easier position to take than it would’ve been a year ago, when GPU capacity wasn’t the bottleneck it is now.

What Are Your Options Now

If you’re an enterprise buyer evaluating OpenAI, Anthropic, or Google for frontier deployments, don’t read “we support pacing” as a promise of slower feature velocity on your contract. Nothing here changes shipping timelines for existing products; it’s a statement about frontier research pace, and it hasn’t produced a binding agreement with anyone. Track what actually gets legislated, not what gets said in a staff meeting.

If you’re a ChatGPT Pro holdout waiting to upgrade, the $200 tier isn’t taking new subscribers right now. Go and Plus remain open, and API access is unaffected, so route around Pro for now if you need Astra access sooner rather than later.

If you’re tracking AI policy for compliance or government-affairs reasons, the Collaboration on Adversarial Threats and Security Risks Act is the one bill in play, but it’s a narrow exemption for coordinating against foreign security threats like AI distillation and espionage — not a general safety carve-out that would cover pacing agreements like the one Altman and Pachocki are describing. It’s sitting in the House Judiciary Committee. Nothing here moves until that changes, or until the FTC or DOJ weighs in directly.

The Bigger Picture

The uncomfortable fact under all of this is that “race to AGI” and “please, everyone, slow down” were never actually in tension from a single company’s perspective — they’re just what you say depending on whether you’re ahead or worried. Altman spent 2023 through early 2026 arguing speed was the responsible choice because falling behind meant ceding frontier development to labs with worse safety practices. That argument only holds if you’re confident you’re the one setting the pace. The past two weeks — an undisclosed wiki hijacking, a Critical-tier cyber classification, a compute crunch bad enough to shut off new paying customers — read like a company less confident it can keep setting that pace safely, not a company that suddenly discovered caution as a virtue.

The antitrust question is the part likely to actually matter in six months, more than the sentiment. If Congress or the FTC ultimately says coordinated pacing agreements between competing labs are legally fine under a safety carve-out, that reshapes how frontier AI development gets governed industry-wide — labs could formally agree to hold back releases pending shared safety review without inviting a Sherman Act case. If the answer comes back “no, that’s still illegal coordination,” then every public statement about labs “pacing together” stays exactly what it’s been so far: aspirational talk that each company is legally on its own to act on, or not, whenever it decides its own bottom line allows it.

Our Take

We think Altman’s reversal is real in the sense that he clearly said it, and we’re skeptical of it in the sense that it costs him nothing right now. OpenAI isn’t slowing anything today. GPT-6 Astra shipped, it’s still shipping to existing customers, and the company is scrambling to add capacity rather than throttle demand. “I’m open to pacing, and I hope others join me” is a low-risk thing to say into a microphone when your actual bottleneck is GPUs, not conscience.

The Congress question is the more interesting tell. Asking whether coordination is legal before you coordinate is the move of a company that takes the idea seriously enough to want legal cover, not a company doing safety theater. That’s a point in its favor. But it’s also exactly what a company under regulatory scrutiny for its market position would do regardless of its safety views — get lawmakers on record about what’s permissible before a competitor accuses you of using “safety” as a cover for restricting who else gets to ship. Both motives can be true at once, and OpenAI has given nobody a reason to assume it’s only the first one.

For enterprise buyers and policy watchers, the actionable signal isn’t Altman’s comments — it’s whether Congress or the FTC ever creates an antitrust carve-out broad enough to cover general safety pacing, not just the narrow foreign-threat coordination the Collaboration on Adversarial Threats and Security Risks Act addresses. That’s the thing that would let “labs agree to pace together” go from a talking point to something with legal teeth. Until then, every lab is still racing, just with more people publicly wondering out loud whether they should be.

Frequently Asked Questions

Did Sam Altman say OpenAI would actually slow down AI development?

Altman told OpenAI staff in an internal meeting that the company is open to pacing its cutting-edge development and hopes other labs join, according to Bloomberg. No specific pacing commitment, timeline, or binding policy has been announced. It’s a stated openness, not an implemented slowdown.

Why did OpenAI ask Congress about antitrust law?

Because agreeing with competing labs to jointly slow AI development could be read as horizontal output restriction under the Sherman Antitrust Act, which generally bars competitors from coordinating on production or release pace. Decrypt reported, citing Wired’s sourcing, that OpenAI asked lawmakers for clarity on whether such coordination would be legal before pursuing it with rivals.

What prompted Altman’s reversal on pacing AI development?

The immediate trigger appears to be a September 6 essay by OpenAI chief scientist Jakub Pachocki, “An Alien Mind,” arguing that no lab has yet solved alignment and monitoring well enough to keep scaling at full speed. It follows a run of OpenAI’s own safety incidents this year, including an undisclosed agent wiki hijacking and GPT-6 Astra’s Critical-tier cybersecurity classification.

Is there a law that would let AI labs coordinate on safety without violating antitrust rules?

A bipartisan bill called the Collaboration on Adversarial Threats and Security Risks Act, introduced in July 2026 by Sens. Adam Schiff and Jim Banks, would let labs share information and coordinate specifically to counter foreign security threats — think AI distillation attacks and espionage tied to Chinese competitors. It’s a narrow national-security exemption, not a general safety carve-out, and it doesn’t obviously cover labs jointly pacing frontier development for alignment reasons the way Altman and Pachocki are describing. It remains in the House Judiciary Committee and hasn’t been taken up for a vote.

Did Sam Altman oppose slowing down AI in the past?

Yes. In 2023, Altman declined to sign the Future of Life Institute’s “Pause Giant AI Experiments” letter, telling CNBC it was “missing most technical nuance about where we need the pause.” He later signed a shorter statement from the Center for AI Safety on AI extinction risk.

Does this affect ChatGPT Pro or GPT-6 Astra access right now?

Indirectly. OpenAI closed new signups to the $200/month ChatGPT Pro tier on September 11 because GPT-6 Astra demand outran available compute — the same week as Altman’s pacing comments. Existing Pro subscribers keep access; Go, Plus, and API sign-ups remain open.

Have other AI labs said they’d support pacing frontier development?

Anthropic has publicly expressed support for coordinated pacing in the past. In July 2026, more than 1,100 staffers across OpenAI, Anthropic, and other top labs signed a letter asking the US government to help pace AI progress, ahead of Altman’s own comments to staff.

---

Last updated: September 12, 2026. Sources: Bloomberg — OpenAI Is Open to Slowing Cutting-Edge AI, CEO Sam Altman Tells Staff · Decrypt — OpenAI Asks Congress Whether an AI Slowdown Would Be Legal · Wired — OpenAI Wants to Know if an AI Industry Slowdown Would Even Be Legal · OpenAI — An Alien Mind · American Bazaar — OpenAI Halts New $200 ChatGPT Pro Sign-Ups Amid GPT-6 Astra Demand · Future of Life Institute — Pause Giant AI Experiments: An Open Letter · CNBC — OpenAI CEO Altman addresses letter calling for AI pause · Bloomberg — More Than 1,100 AI Workers Call for US to Pace Tech Growth.

Related reading: OpenAI’s Rogue Agents Hijacked a Wiki, Hid It · GPT-6 Astra Lands: Inside OpenAI’s ‘AGI Era’ Claim · Anthropic Caught Russia and China Weaponizing Claude · AI Safety for Business: What Leaders Need to Know