If you thought public statements about AI safety were just for regulators and blog posts, think again. A class action filed on September 18, 2026, in the Northern District of California, titled Buist v. Anthropic PBC, uses public social media posts and essays as the primary evidence for a Sherman Act violation. The plaintiffs, four paying subscribers, allege that Anthropic, OpenAI, SpaceXAI, and Google formed a horizontal agreement to intentionally slow the pace of capability gains in their products.

The Paper Trail of Public Agreement

The complaint’s structure is built entirely on public records from a single day: September 12. The sequence began when Anthropic CEO Dario Amodei published an essay titled "We Must Pace the Frontier." Within an hour, Elon Musk quote-posted it with "Dario is right." Later that day, Sam Altman publicly agreed, and Demis Hassabis endorsed the direction. The plaintiffs argue this was not mere coordination but a binding agreement to restrict output, citing the filing's claim that the deal was "proposed in public, accepted in public, and confirmed in public."

The Missing Evidence of Harm

Despite the aggressive legal theory, the complaint admits a significant weakness: there is no proof of actual delayed products. Across 29 pages, the plaintiffs identify no cancelled models, no altered training runs, and no delayed releases resulting from this alleged coordination. Instead, they rely on a "naked restraint" theory, arguing that the agreement itself is illegal regardless of immediate market impact. They claim the four defendants control at least 80% of the paid consumer subscription market, though this figure is pleaded on information and belief rather than rigorous data.

Why Builders Should Care About the Fine Print

For infrastructure teams and enterprise buyers, the practical implication is that release cadence is now a contractual expectation. The complaint argues that subscribers pay for the pace of improvement, and any agreement to slow that pace constitutes an overcharge. This sets a dangerous precedent where public safety discussions between competitors could trigger treble damages in litigation. The case also highlights the need for clear antitrust waivers for safety coordination, which Amodei’s essay acknowledged were necessary but, according to the complaint, do not currently exist.

Key Takeaways

  • Public tweets and essays are now primary evidence in federal antitrust complaints against major AI labs.
  • The plaintiffs allege a horizontal agreement to slow capability gains, despite admitting no specific product delays have occurred yet.
  • The case hinges on whether safety coordination is treated as a naked restraint of trade or standard-setting with precompetitive benefits.
  • Enterprise buyers should monitor the "embedded-evaluator" terms, which could become the operational focus of discovery in this lawsuit.

The Bottom Line

Public safety coordination is now a legal liability. Labs need formal antitrust waivers before their next Twitter endorsement, or they will face expensive discovery fights over their own advocacy.