The Trump administration is turning up the heat on Meta, pushing the company to submit its AI models for federal safety reviews before releasing them to the public. It’s a notable ask from an administration that, not long ago, was busy tearing down Biden-era AI guardrails in the name of innovation.

Here’s the thing: Meta is increasingly the odd one out. OpenAI, Google, Microsoft, and xAI have already committed to or begun participating in the review process. Meta, the company behind the open-source Llama family of models, has so far not joined the club.

What the executive order actually requires

President Trump signed an executive order on June 2, 2026, establishing a voluntary review framework for advanced AI models. The process is straightforward in concept: companies submit their most powerful AI models to federal agencies for a 30-day review period before making them publicly available.

The reviews carry a cybersecurity focus, which makes sense given what prompted them. Discussions around the policy began in May 2026, following briefings that reportedly highlighted concerning capabilities in Anthropic’s Mythos model. That model demonstrated an ability to exploit advanced vulnerabilities, the kind of capability that tends to get national security officials’ attention very quickly.

Meta’s complicated position

Meta occupies a unique and somewhat awkward spot in this debate. The company’s Llama models are open-source, meaning once they’re released, the weights are publicly available for anyone to download, modify, and deploy. That’s fundamentally different from the closed-model approach taken by OpenAI or Google, where access is mediated through APIs and usage policies.

This distinction matters enormously for the review process. A 30-day hold on a closed API product is an inconvenience. A 30-day hold on an open-source release is a philosophical challenge, because the entire value proposition of open-source AI is that it’s freely available and openly distributed.

Meta’s Llama models already received US government approval for agency use back in September 2025. That approval, though, was about government adoption of existing models, not about pre-release scrutiny of new ones.

A policy U-turn with extra steps

Early in his term, Trump rescinded previous Biden-era directives aimed at enhancing AI safety. Initial discussions started in May 2026, and the executive order followed about two weeks later. OpenAI CEO Sam Altman has publicly supported what he describes as a balanced voluntary safety framework for AI models.

What this means for investors

For Meta specifically, the pressure creates a strategic tension. If the company complies, it introduces a delay mechanism into its open-source release cycle that competitors like Mistral or other international players don’t face. If it doesn’t comply, it risks becoming a political target in an environment where bipartisan support for some form of AI oversight is growing.

One thing worth tracking: whether this pressure eventually forces Meta to reconsider the degree of openness in its AI releases. A world where government pre-screening becomes the norm is a world that structurally favors closed-source AI development, and that reshuffling of competitive advantages could have real implications for valuations across the sector.