The U.S. government and Anthropic hashed out an agreement to allow the launch of its most powerful model Mythos only to 100 pre-approved U.S. government agencies and companies.
Anthropic’s Fable 5 is still shelved from public use.
The Trump Administration’s voluntary collaborative review of powerful AI models appears to be expected by the U.S. government before they are launched publicly. This review may be voluntary in the sense that no law requires it. But if an AI company does launch a new frontier model, it could face an export control as Anthropic faced with Fable 5. That forced Anthropic to shut down public access to Fable 5.
Given how powerful these frontier models are becoming, having a layer of government review, especially for cybersecurity and national security risks, is sensible. No one wants AI to enable malevolent actors to penetrate U.S. defense systems or to bring down the global financial system. Yet, the informal process is opaque and thus far behind closed doors.
I anticipate that, in the near future, a more formalized process for all U.S. AI companies will be developed.
