The U.S. government has pressed Meta to hand over its latest AI models for a formal safety review. The request arrives after the Biden administration created the Center for AI Standards and Innovation, staffed by technical experts tasked with vetting frontier AI. Meta remains the only major developer that has not voluntarily submitted its models, while rivals OpenAI, Anthropic, Google, xAI and Microsoft already provide early access.

Meta’s newest model, Muse Spark, launched in April, offers an “Instant” mode and a “Thinking” mode that pauses to reason before answering. Government officials have been emailing the company to secure a review, citing the June 2 executive order that gives agencies until July’s end to craft a 30‑day evaluation window. OpenAI and Anthropic are already testing unreleased versions with the same agency.

If Meta signs an agreement, regulators will gain its first‑hand insight into Muse Spark’s reasoning layer, potentially shaping compliance standards for commercial AI. The push reflects growing bipartisan concern that unchecked models could expose national‑security vulnerabilities, as seen when the administration forced Anthropic to block foreign access to its Mythos 5 and Fable 5 systems. A formal review would place Meta under the same scrutiny as its competitors.