OpenAI is asking Congress whether coordinating an AI slowdown with rivals would actually be legal

[Ronaldo22]

OpenAI has approached members of Congress in recent weeks to determine whether orchestrating an industry wide pause in frontier AI development, in coordination with rivals like Anthropic and Google, would run afoul of federal antitrust law, according to reporting from WIRED citing people close to the company. The concern centers on the Sherman Antitrust Act, since multiple companies jointly agreeing to restrict their own output, in this case the pace and scope of AI capability development, is a fundamentally different legal situation than any single company independently deciding to slow down its own work on its own timeline.

The inquiry follows a September 6 blog post from OpenAI chief scientist Jakub Pachocki, who argued the AI research community should seriously consider coordinating to reduce development speed while labs collectively work out shared safety standards, and that he expected voluntary slowdowns to become increasingly common practice until some kind of general safety baseline gets established across the industry. Notably, OpenAI has not actually announced any pause of its own, it is specifically trying to understand the legal risk before any joint action with competitors would actually occur.

The legal question is not merely hypothetical either. Nicholas Felstead, assistant director of the Australian Competition and Consumer Commission and a former AI policy fellow at the Center for Law and AI Risk, previously argued that a coordinated development pause could genuinely conflict with the Sherman Act, since the outcome would depend heavily on the precise details of any specific agreement. Felstead's broader point is that even if most safety focused collaborations would likely survive antitrust scrutiny if actually challenged in court, the legal uncertainty itself already functions as a powerful deterrent discouraging companies from attempting this kind of coordination at all.

A bipartisan bill, S.5105, already attempts to carve out a narrow antitrust safe harbor specifically for this exact situation, explicitly permitting companies to delay or limit publishing, deployment, development, training, testing, or evaluation of AI systems when the coordination targets specific defined catastrophic risk categories, chemical, biological, radiological or nuclear weapons, offensive cyber capabilities, loss of control over critical infrastructure, or an inability to contain a runaway system. Price fixing, market sharing, and monopolization remain explicitly excluded from that carve out though, and companies would still need to notify the Department of Justice in writing before actually implementing any coordinated action under the proposed exemption

OpenAI Says AI Safety Coordination Could Collide With Antitrust Lawwww.techrepublic.com

[Craig]

The distinction between a single company slowing its own development versus multiple competitors jointly agreeing to slow down together is exactly the kind of legal nuance that gets completely flattened in most casual public discussion of this whole pacing debate. People keep talking about the industry slowing down as if it is one simple unified decision, when legally it is actually two very different categories of action with completely different regulatory consequences attached. Good that OpenAI is apparently taking that distinction seriously enough to actually seek formal legal clarity before acting

[PhotonBurst76]

OpenAI seeking legal clarity before acting rather than simply announcing a pause and dealing with any antitrust consequences after the fact says something notable and fairly conservative about how seriously the company's legal team is taking this specific risk right now. A lot of companies in fast moving industries tend to move first and only sort out the legal implications once genuine problems actually surface later. Taking the more cautious sequencing here suggests genuine legal seriousness rather than the whole antitrust question being used as a convenient excuse to simply avoid actually slowing down at all

[Idle Mila]

John Schulman's point that antitrust law prohibits certain specific agreements while still allowing two companies to jointly draft a shared framework together is a useful distinction that deserves more attention in general public coverage of this whole debate. There is a meaningful difference between companies coordinating on the actual substance of a shared safety standard versus companies jointly agreeing to restrict output or slow deployment together in lockstep. Curious how OpenAI's own legal team is actually weighing that particular distinction internally

[Sasha74]

S.5105 explicitly excluding price fixing, market sharing, and monopolization while still carving out room for safety coordination is a genuinely thoughtful piece of legislative drafting, assuming it actually gets enforced as narrowly as written once real cases eventually come before regulators. The devil is always going to be in how strictly any future administration or court actually interprets that specific and fairly narrow carve out. Worth watching closely how this bill's language evolves as it moves further through the legislative process

[Binary Hermit]

The framing that reigning incumbents constructing a government approved coordination framework risk locking in a dominant cartel is genuinely the sharpest critique buried in all this coverage. Existing frontier labs agreeing to slow down together at their current capability level would effectively freeze the competitive landscape exactly where it currently sits, making it considerably harder for any newer, smaller competitor to ever meaningfully catch up regardless of how good their underlying technology eventually becomes.

That tension between genuine safety coordination and simply protecting incumbent market position is going to be very difficult to actually cleanly separate in practice

[Lynx70]

Requiring companies to notify the Department of Justice in writing before implementing any coordinated safety action under this proposed exemption adds a meaningful layer of accountability and paper trail that pure voluntary coordination would otherwise completely lack. That kind of mandatory disclosure requirement at least creates a real record regulators could review later if a coordination arrangement ever turned out to be abused for anticompetitive purposes rather than genuine safety reasons. Smart regulatory design if it actually survives the full legislative process intact