Context:

Efforts to launch a federal AI safety and testing framework are stalled by internecine disagreements within the Trump administration and a looming timeline around the China summit, delaying concrete regulatory action after the surge of frontier models like Mythos. Key officials clash over how aggressively to regulate or test new systems, with mixed messages about whether to create new agencies or rely on existing processes. The debate is further complicated by international dynamics, as Europe seeks access to advanced models for security testing, and by practical steps such as rapid shifts in who should oversee testing within government. The situation signals likely continued ambiguity in US AI policymaking, with decisions possibly tied to the outcome of the China meetings and evolving industry developments the near term. Expect further jockeying for clarity on governance, with executive actions viewed as a potential interim path while Congress drags toward a longer-term solution.

Dive Deeper:

• A dispute among White House aides, industry officials, and lawmakers is slowing momentum for a federal AI safety review framework, with a sense that administration-wide cohesion is lacking just as frontiers like Mythos reshape the landscape.

• Officials have floated options ranging from a centralized regulatory approach to reassuring cautions about deploying tests and results, but public endorsements or concrete steps remain elusive, reflecting internal disagreement over risk, speed, and scope.

• The debate is set against a time pressure created by upcoming talks at the China summit, with some sources suggesting the administration will defer major policy moves until after the meeting to gauge international responses.

• Concerns about national security and cybersecurity are driving calls to harden systems against rapidly advancing models, even as some White House voices push back on creating a sweeping new bureaucracy or FDA-like approval process for AI.

• There is an ongoing tug-of-war over whether frontier AI testing should occur under civilian agencies like the Department of Commerce or within national security-leaning offices, highlighted by the temporary removal of a CAISI testing page after it was publicized.

• Key figures cited include National Economic Council Director Kevin Hassett, White House Chief of Staff Susie Wiles, and PCAST co-chair David Sacks, who offer divergent views on pacing and structure, complicating consensus.

• Industry observers underscore the need for clear, forward-looking guidance to avoid recreating the same ad hoc processes with each new model breakthrough, signaling waiting momentum until a stable framework emerges.