The US government is putting pressure on Meta to submit its AI models for safety and security evaluation. According to Engadget, citing a report from The New York Times, Meta is the only major AI company that hasn’t yet agreed to hand over its models for government review. Federal authorities want to assess what those models can do and identify any potential vulnerabilities before they reach the public.
The request has reportedly been coming through email, and Meta hasn’t outright refused. Spokesperson Francis Brennan said the company supports the Trump administration’s goal of advancing US leadership in “robust and secure frontier AI” and added: “While we are working through the details, we hope to sign the agreement soon.”
This puts Meta in an awkward spot compared to its rivals. OpenAI and Anthropic are already working directly with the government to test unreleased models. Google, xAI, and Microsoft have all agreed to give the Center for AI Standards and Innovation early access to their new systems. That agency, which was created under the Biden administration and staffed with technical experts, is now led by Commerce Secretary Howard Lutnick.
The broader context here matters. On June 2, President Trump signed an executive order to build a formal framework for evaluating AI releases. The government has until the end of July to put that review process together. The goal is for companies to give authorities up to 30 days to assess their models before making them public. Even without a formal process in place, most major players have been sharing models voluntarily for months. Meta has not.
Meta’s latest model is Muse Spark, which it released in April. It has two modes:
- Instant: Standard responses for quick queries
- Thinking: A reasoning mode that takes a bit more time to work through a prompt before answering
Muse Spark isn’t considered a frontier model in the same league as competitors’ top offerings, but the government’s scrutiny isn’t limited to the most powerful systems. Federal authorities have been watching the whole AI industry more closely lately.
That scrutiny showed up clearly in mid-June, when the government ordered Anthropic to cut off access to its Mythos 5 and Fable 5 models for all foreign nationals, citing national security concerns. Anthropic responded by blocking everyone’s access while it worked out how to comply. Mythos is Anthropic’s specialist cybersecurity model, available only to partners in its Project Glasswing program. Fable 5 was built to bring many of those capabilities to the general public and, according to Anthropic, outperformed any model the company had previously released.
The episode signals just how seriously the government is taking AI security right now. For Meta, the pressure to fall in line with its peers is only likely to grow as the July deadline for a formal review process approaches.
