Something strange is happening in artificial intelligence. The two leading labs — OpenAI and Anthropic — are finding themselves in the exact same predicament, and the old narrative of competing against each other doesn’t really apply anymore.
Two weeks after the U.S. government pulled Anthropic’s Fable and Mythos models from public release, OpenAI’s GPT-5.6 appears headed for identical limbo. The Information reported that the model is being released only in limited preview, with the government approving access “customer by customer.” Sam Altman reportedly projected a couple of weeks in review. Mythos has been in preview for months with no end in sight.
The core problem is that the U.S. government is building a haphazard approval process for frontier AI models, and nobody knows the rules yet. Governments can absolutely test models before release. Medicines go through this process all the time. But as policy researcher Dean Ball pointed out, it’s not obvious what safety checks would actually satisfy regulators. The U.S. government lacks the technical capacity for that kind of deep evaluation. And regulators haven’t clearly articulated which specific risks they’re trying to guard against in the first place.
There are real concerns underneath the bureaucratic mess. AI tools are revolutionizing cybersecurity in visible ways. Biorisk and alignment are genuine issues. But restricting model releases can’t be the whole answer ’ that only limits what’s available to the public while doing nothing about models that already exist.
The best path forward, as outlined by AI policy researchers, requires the industry to work together. That means trusting independent groups to guide the process, lining up behind the least-bad regulatory options, and treating safety as a shared interest rather than a competitive weapon.
For people with billions riding on one company or another, that’s a tough sell. But the political consequences of these models are too big for any single lab to handle alone. Whether the industry figures that out will define the next chapter of AI development.
