The consensus read on the AI slowdown debate treats it as a live question, an open negotiation where outcomes remain genuinely uncertain. The structure of Dario Amodei’s own plan, examined alongside the incentives of every named participant, suggests AI slowdown coordination failure is not a risk but a near-certainty.
Amodei has called publicly for AI companies to slow model development, drawing nominal support from Sam Altman and Elon Musk. The optics are striking: rivals converging on a shared concern. But alignment in rhetoric has never produced alignment in behaviour, particularly when the financial stakes are this large and no binding mechanism exists.
What Amodei’s Plan Actually Requires
The specifics of the proposal, as Reuters reported, reveal how much the plan depends on voluntary goodwill from competitors with no obligation to provide it. Amodei’s three-step framework calls for embedded independent evaluators with employee-like access to verify safety practices at frontier AI firms, coordination among those same firms to set safety standards and limit unchecked development, and international cooperation to manage AI risks.
Step one assumes frontier labs will open themselves to outside scrutiny. Step two assumes competitors will coordinate rather than exploit each other’s restraint. Step three assumes international cooperation in a geopolitical environment where China has shown no inclination to join any such arrangement. Each step is individually optimistic. Together, they require a level of institutional trust that does not exist in this industry.
There is also an antitrust dimension that has received little attention. According to the Associated Press, one component of Amodei’s plan asks the US government to potentially issue waivers allowing AI companies to coordinate on safety standards without violating antitrust law. Read that carefully: the plan requires government permission for competitors to talk to each other at all. And the current administration has already rejected any pause in AI development. So the legal scaffolding the plan needs is being requested from an administration that has explicitly refused the premise.
The Threat Amodei Is Describing Is Real Enough
To be clear, the underlying concern driving this call is not confected. Yahoo News reported that Amodei warned swarms of rogue AI agents could, within 6 to 12 months, be capable of taking over the entire internet via a persistent botnet, potentially causing hundreds of billions of dollars in damage. That is a concrete and specific warning, not a vague civilisational hand-wave. Whatever one thinks of the probability, the mechanism is at least coherent.
The problem is that a genuine threat does not automatically produce a coordinated response. History is not short of examples where everyone acknowledged a shared risk and acted in their own narrow interest anyway. The AI slowdown debate has exactly this structure, and Amodei’s plan, to his credit, implicitly acknowledges it by building in enforcement mechanisms. The trouble is those mechanisms require the cooperation of parties who benefit from non-cooperation.
Musk, meanwhile, has made the economic stakes explicit. His stated projection that AI will grow the global economy by 20 to 30%, amounting to around $20 to $30 trillion a year, frames this as a prize too large to walk away from unilaterally. Whether or not one believes that figure, it explains the strategic logic perfectly. Nobody volunteers to slow down when the competitor who doesn’t slow down captures the upside.
The ‘you first’ dynamic here is not a failure of courage or good intentions. It is a straightforward collective action problem with no authority capable of resolving it. China has given no indication it intends to participate in any slowdown framework. US firms can therefore point to China as justification for continued development, regardless of whether any superficial agreement is reached domestically. And if an agreement is reached, the incentive to defect quietly while publicly affirming compliance is obvious to everyone involved.
The AI slowdown coordination failure scenario does not require bad actors. It only requires rational ones. Amodei’s plan is the most structured proposal on the table, and it still depends on the antitrust waivers, voluntary competitor access, and international goodwill that the current environment makes unlikely. The proposal is worth taking seriously as a diagnosis. As a remedy, the gap between what it requires and what the political and commercial landscape can deliver is too wide to ignore.
