While loudly raising funds and going public, they propose to pause AI development: Anthropic is just playing safety public relations

According to Beating Monitoring, in its report Building AI Itself, Anthropic proposes establishing, by reference to global nuclear arms control—such as the Intermediate-Range Nuclear Forces (INF) Treaty— a “verifiable cooperative pause mechanism.”

The report explicitly opposes unilateral suspension of R&D, and emphasizes that only on the condition of globally verifiable verification can leading labs apply the brakes in sync.

However, cooperative braking is fundamentally impossible in technical terms. Nuclear arms control can rely on satellites to watch missile silos, but graphics cards are small, easy to conceal, and can be used for distributed computing.

To verify whether other parties are secretly training, the only way is to carry out intrusive monitoring of all GPU chips, power grids, and data centers worldwide.

Comprehensive monitoring is not only politically unworkable, but also, from a technical standpoint, equivalent to advocating computational-power authoritarianism. Against the backdrop that technical verification cannot be implemented, the proposal for a cooperative pause is more like a “safety disclaimer card.”

Right on the eve of releasing the report, Anthropic just completed a $65 billion funding round at a valuation close to one trillion and filed for a listing application. By setting the conditions for “the pause” as an impossible form of global verification, Anthropic paves the way for its own indefinite training while also shifting responsibility for not being able to stop onto a regulatory mechanism that has yet to be established.