Meta pressed to join Trump AI security review framework
In brief
- Meta remains sole major US AI company without federal model evaluation agreement
- Trump administration established voluntary review framework June 2, 2026, led by NSA, CISA, NIST
- OpenAI, Anthropic, Google, xAI, Microsoft already signed onto voluntary review process
- 30-day government review creates friction in Meta's open-source Llama model strategy
- Federal clearance could provide credibility advantage in sensitive sectors and enterprise deployments
The voluntary framework
President Trump signed an executive order establishing the review structure, which gives the federal government up to 30 days to evaluate AI systems before their public release. Three agencies are leading the effort: the NSA, CISA, and NIST, each tasked with developing benchmarks that assess the national security implications of advanced frontier AI models. The reviews are being run through the Center for AI Standards and Innovation, or CAISI. By the end of July 2026, the full review process is expected to be operational.
Rather than imposing binding regulations on AI developers, the White House is opting for a cooperative model. This approach mirrors the Biden administration's 2023 push for voluntary commitments from AI companies, though the Trump version includes explicit government evaluation timelines and agency coordination.
Meta's resistance
Meta has, so far, declined to volunteer. The company has built its Llama model family on the premise that open access drives innovation and adoption. Submitting models for a 30-day government review before release introduces friction into that process, a delay that conflicts with Meta's strategy of rapid iteration and broad availability.
The tension runs deeper than operational inconvenience. Meta's open-source models are already viewed with skepticism by some in the national security community. A federal review stamp could reshape how enterprise customers and government contractors evaluate AI vendors.
The credibility gap
If every major AI lab except one is submitting to government reviews, enterprise customers and government contractors will eventually start asking which vendors have been cleared and which haven't. Models that have undergone federal evaluation might be perceived as safer, more trustworthy, and more suitable for deployment in sensitive industries.
Meta now faces a choice: maintain its open-source velocity at the cost of perceived security credibility, or accept the 30-day friction and gain federal clearance. The administration's pressure suggests this is not a casual request.
