Meta is the only major AI company refusing to submit its models for federal government review. The Trump administration issued an executive order on June 2 establishing a framework for evaluating AI systems before public release. OpenAI, Anthropic, Google, Microsoft, and xAI have all agreed to participate. Meta has not.
The government gave companies until the end of July to develop a review process. The goal is for the government to have up to 30 days to evaluate new AI models before companies release them publicly. The focus is on identifying vulnerabilities and assessing capabilities, particularly around threats to critical infrastructure.
Meta’s position puts the company in isolation. Not for ideological reasons. It’s practical business strategy. Mark Zuckerberg doesn’t want government officials examining Meta’s proprietary AI systems. He doesn’t want regulators finding reasons to impose restrictions.
But holding out creates other problems. The administration is pushing. The New York Times reported that officials are urging Meta to join. Publicly resisting makes Meta look like the bad actor. The company cares about public perception. Regulators care about who cooperates.
The executive order itself is measured. Thirty days isn’t unreasonable. Companies get time to respond to feedback. The government promises confidentiality for trade secrets. But it’s a precedent. Once government review starts, it normalizes government oversight of AI.
That’s probably why Meta is resisting. Zuckerberg has been burned by regulation before. Facebook faced FTC oversight over privacy. The company spent years dealing with government inquiries. Meta doesn’t want to go down that road again with AI.
But refusing cooperation has costs. It signals non-compliance to regulators. It creates suspicion. Why is Meta the only company not participating? Is the company hiding something? Are its safety practices inferior?
The other companies made a calculation. Cooperating looks good. It demonstrates responsibility. It gives the government information without giving up control. By being first to cooperate, OpenAI and Anthropic position themselves as the responsible actors. By not cooperating, Meta becomes the outlier.
There’s also a competitive angle. If the government gets early access to Meta’s competitors’ models, regulators understand their capabilities. Meta’s secrecy might actually put the company at a disadvantage. Regulators develop rules based on what they can see. If they only see OpenAI and Anthropic’s systems, the rules will be written around those systems.
Meta’s position is weakening. The company spent months dealing with regulatory pressure over content moderation and privacy. Adding AI regulation resistance to that list makes things harder. Advertisers, users, and regulators all prefer companies that appear to play by rules.
The deadline is end of July. Meta could still join. The company could submit a model for review. It would be presented as new thinking, not capitulation. By then, the framework will be clearer anyway.
Stubbornness on regulation usually costs companies more than compliance does. Meta should probably reconsider.
