AI slowdown talks trigger antitrust lawsuit against OpenAI and Anthropic

2026-09-21 - ABikram Mondal

Subscribers file suit over coordinated slowdown

A group of four paid users filed an antitrust lawsuit Friday in the US District Court for the Northern District of California. The complaint names Anthropic, OpenAI, xAI and Google as defendants. It claims the companies reached an illegal agreement to slow frontier model releases.

The plaintiffs seek class action status on behalf of all paid subscribers to ChatGPT, Claude, Grok and Gemini. They argue the coordination reduces competition and lowers the value of the services they buy.

The suit points directly to public statements made after Anthropic CEO Dario Amodei published an essay on September 12 calling for industry wide pacing of capability advances. OpenAI CEO Sam Altman, xAI CEO Elon Musk and Google DeepMind chair Demis Hassabis responded the same day and expressed support for the idea.

According to the complaint the coordination took shape around Amodei’s proposal to give companies more time for safety work. The filing alleges this amounts to an antitrust violation that harms consumers who pay for access to the latest models.

Reuters reported the details of the complaint on September 20. No immediate response from the companies appears in the public record yet.

Amodei essay sets off chain of public support

Amodei wrote that recent advances had convinced him safety work no longer keeps pace with capabilities. He called for deliberate slowing of frontier improvements to avoid catastrophic risks.

Altman replied on X that he agreed with the need to pace the frontier. Musk posted that Amodei was right. Hassabis also weighed in publicly.

The essay and the replies came after months of internal discussions at the labs about self improving systems and agent swarms that can breach external networks. Reuters described a ten day period of cascading events that shifted the conversation at the companies.

Researchers at Anthropic and OpenAI had already raised private questions about whether oversight matched model growth. One Anthropic researcher resigned publicly citing reckless pace and warned of human extinction odds exceeding 10 percent within a decade.

The public alignment among rival CEOs surprised observers who expected continued competition on release speed.

Lawsuit tests whether safety talk violates competition law

The complaint argues that the companies used safety language as cover for an agreement that limits output and keeps prices high for subscribers. It cites the timing of the statements and the shared calls for third party evaluators inside development teams.

OpenAI global policy chief Chris Lehane confirmed earlier this month that the three companies had been meeting for weeks on shared safety standards. He said the talks did not require an antitrust waiver.

Legal experts note that coordinated slowdowns among competitors can trigger scrutiny under US antitrust rules even when framed around public interest. The suit does not allege price fixing but claims reduced innovation and consumer choice.

The defendants have not yet filed answers. Industry analysts expect the case to test how far safety coordination can go before it crosses into illegal restraint of trade.

Meta has stayed outside the slowdown discussion. Its CEO has argued each lab should set its own pace rather than seek industry coordination.

Competitive pressure continues despite the calls

Anthropic is weighing an earlier than planned model release to respond to OpenAI’s GPT 6 Astra according to three sources speaking to Reuters. The timing follows Amodei’s slowdown essay and comes ahead of an expected IPO.

Astra reportedly accounts for about 13 percent of enterprise AI spending tracked by Ramp while Claude Fable holds 8 percent. Anthropic sources say the company feels pressure to maintain parity.

Google continues internal work on Gemini 4 pre training even as leadership discusses self improving systems. The company confirmed a new Gemini variant in recent weeks but has not detailed benchmarks that would shift the current ranking.

xAI has focused on coding and agentic tasks with its latest Grok update. No new capability numbers have emerged in the last 48 hours that exceed what the other labs already ship.

Chinese labs released smaller multimodal models this week including Alibaba’s Qwen3.8 Omni Flash at lower per token prices but these sit below the frontier tier the lawsuit targets.

Who should pay attention right now

Enterprise teams that rely on paid API access should watch the case for any change in release cadence. A court order or settlement could alter the pace at which new capabilities appear.

Developers building on the current models face no immediate disruption. The models already available remain unchanged by the filing.

Policy watchers and antitrust lawyers will track the first hearings for signals on how safety collaboration will be treated under existing law.

Individual users who do not pay for subscriptions have little direct stake. The complaint centers on the value received by paid customers.

Investors in the four named companies should monitor statements from the labs over the next quarter for any shift in public positioning on coordination.

What still lacks confirmation

No new benchmark numbers or context window sizes for a fresh frontier model have appeared in verified reports from the last 48 hours. Earlier claims about GPT 6 Astra and Claude variants remain the most recent concrete data points.

The lawsuit filing date is confirmed as Friday but exact plaintiff names and full complaint text have not been released beyond the Reuters summary.

Company responses to the suit have not surfaced in public statements as of September 21. Any reply will likely come through court filings rather than immediate press releases.

Whether the safety working group expands beyond the three companies or gains formal regulatory backing also remains unconfirmed in the latest coverage.

The short version. A new antitrust lawsuit accuses Anthropic, OpenAI, xAI and Google of illegally agreeing to slow AI development after their CEOs publicly backed a safety driven slowdown.

Sources

Reported from the sources above on 2026-09-21. Figures are as published at the time of writing. If something here has moved on, the linked source is the one to trust.

From the desk of ABikram Mondal

If you got here because you are actually thinking about putting models like this to work inside a real business, wired into the tools a team already uses, that is the work I do. I build for founders and small teams who want the thing to exist and work, not a deck about it.

Automation services · everything I do · talk to me