AI safety coordination needs a public protocol, not an informal rival pact
A lawsuit alleges leading AI companies coordinated a slowdown after public calls for pacing. Whatever the case’s merits, shared safety action needs a narrow mandate, transparent evidence and independent oversight.
What happened
A federal antitrust lawsuit accused Anthropic, OpenAI, SpaceXAI and Google of an illegal agreement to slow AI development.
Why it matters
Frontier labs may need to coordinate on safety, but opaque coordination among competitors can undermine accountability and competition.
The Associated Press reported that a lawsuit filed in the Northern District of California accuses Anthropic, OpenAI, SpaceXAI and Google of illegally agreeing to slow AI development. The complaint draws on public responses to Dario Amodei’s September 12 call to pace frontier progress. The defendants had not immediately responded in the report.
An allegation is not a finding. The article cannot establish that an agreement existed, restrained competition or harmed subscribers. It does expose an operating-design problem: how can rivals coordinate on a genuine shared hazard without turning safety into an opaque market arrangement?
Publish the coordination object
Coordination should be about a narrowly specified control, not prices, customers, output or broad product timing. Define the trigger, evidence threshold, affected capability, maximum duration, review authority and exit condition. Publish the protocol before it is invoked and record each invocation afterward.
An independent body should hold the evidence and decide whether the trigger was met. Firms can submit confidential technical material under a consistent process, but the public should see the rationale, scope and duration. Participation and non-participation should be documented. Customers need to know which service commitments change and what remedies apply.
Separate unilateral duties from collective action
Every lab can act alone on evaluation access, incident reporting, deployment gates and credential controls. Collective action should be reserved for risks that genuinely cannot be managed unilaterally. That distinction prevents companies from withholding ordinary safeguards while waiting for competitors.
The strongest counterargument is that disclosure could reveal dangerous capabilities or make rapid response impossible. A protocol can protect technical details while still publishing the decision rule, authority and aggregate evidence. Emergency action can be temporary, followed by prompt review.
Legal questions require qualified counsel and human domain review; this draft reaches no conclusion on the complaint. For enterprise buyers, the immediate lesson is contractual: require vendors to disclose which external coordination protocols may change access, performance or roadmaps, and what audit trail will follow.
The governance goal is not “coordination” in the abstract. It is a mechanism that makes a shared safety decision bounded, reviewable and distinguishable from a commercial pact.