The US government has been quietly pressing Meta to do what every other major AI company has already agreed to: let federal officials review its most powerful AI models before they go public. OpenAI, Anthropic, Google, Microsoft, and xAI have all reached agreements with the Center for AI Standards and Innovation, a Commerce Department body created to vet AI technology. Meta has not. According to the New York Times, the government has been sending emails asking Meta to participate.
Meta's spokesperson says the company shares the administration's goals and expects to sign an agreement soon. That is unlikely to be the end of the story.
The legal foundation for all of this is an executive order Trump signed on June 2. It asks companies to voluntarily give the government up to 30 days to evaluate their most powerful new AI models for security risks before releasing them. The government has until the end of July to finalize the actual review process. The order was originally drafted with a 90-day review window, but that was cut to 30 days after concern inside the administration that a longer period would slow US companies down relative to Chinese competitors.
This is a notable shift. On his first day in office in 2025, Trump had cancelled a Biden-era requirement for AI companies to share safety tests with the government. What changed the administration's thinking was the emergence of AI models with serious cybersecurity capabilities. The government's concern is that if a powerful AI model has a security flaw that bad actors can exploit, and it reaches millions of users before anyone spots the problem, the damage could be severe.
That concern is not theoretical. Earlier this month, the government forced Anthropic to pull two of its newest models, Fable 5 and Mythos 5, citing a national security vulnerability. The order arrived at 5:21 PM on a Friday. Anthropic had less than 90 minutes to respond. Because the company could not verify who was a foreign national and who was not, it shut down both models for every single customer globally, including businesses that had built workflows around them. The government's stated reason was a method for bypassing the model's safety guardrails that had been shared with officials. Anthropic disputed the severity, saying the vulnerability was minor and already present in other publicly available AI tools.
For businesses using AI tools built on top of any of these major models, the Anthropic episode is the clearest signal yet that government intervention can cut access with almost no warning. If your operations depend on a specific AI tool, the question worth asking is whether there is a backup if that tool disappears for days or weeks.
Meta's situation adds another layer of complexity. When Meta launched Muse Spark in April, it marked a clean break from its previous strategy. For three years, Meta had released its AI models as open-source software, meaning any company or developer anywhere could download and use them for free. Muse Spark is the opposite: fully proprietary, no public access, and even more locked down than competitors like OpenAI and Anthropic. The switch happened after Meta spent $14.3 billion to bring in Alexandr Wang from Scale AI and rebuild its entire AI operation from scratch. The concern that drove the decision was partly that Chinese companies had been using Meta's free models to build their own competitive products without contributing anything back.
The government's push to review Meta's models fits into this context. Muse Spark is a closed, proprietary model powering Meta's apps across more than three billion users. That is a different kind of risk profile than an open-source model that anyone can inspect.
The broader pattern here is that AI is moving from a largely unregulated space into one where governments are actively intervening, but without clear rules or timelines. The current US approach is voluntary and informal. The formal review process is still being written. That gap between informal pressure and formal rules is where things like the Anthropic shutdown happen: abrupt, large in impact, and difficult for businesses to plan around.