Two weeks after the U.S. government pulled Anthropic’s Fable and Mythos models from general release, OpenAI’s next-generation model appears to be facing the same regulatory bottleneck. According to a report from The Information on Thursday, OpenAI’s GPT-5.6 will launch only in a limited preview, with the government approving access “customer by customer” until a broader release is authorized.
If OpenAI CEO Sam Altman’s projection of a “couple of weeks” for the preview phase holds, the disruption may be manageable. But the precedent is worrying: Anthropic’s Mythos has been stuck in preview for months, with no clear path to a general release. Even short review windows could significantly erode the economic returns of expensive new systems, at a time when both labs are under pressure to improve their bottom lines.
Also read: OpenAI Limits GPT-5.6 Rollout After U.S. Government Request, Calls Restrictions a 'Short-Term Step'
A Shared Problem, Not a Competitive Advantage
The instinct within the tech industry has been to frame the situation as a rivalry. Some accuse Anthropic of running a regulatory capture scheme; others accuse OpenAI of cozying up to the Trump administration to freeze out a competitor. These narratives reflect the billions of dollars at stake for the prominent figures backing each lab.
But the core issue transcends corporate competition. Both labs now face the same fundamental challenge: a haphazard government approval process that lacks clear criteria, timelines, or technical capacity. As George Mason University fellow Dean Ball detailed in a recent post, the U.S. government does not have the expertise or infrastructure to conduct the kind of testing that would be required for frontier AI models. More critically, regulators have not articulated what specific risks they are trying to guard against.
Also read: OpenAI, Google, and SpaceX are all building their own AI chips — here's why that matters
The Real Risks Behind the Regulatory Fog
It would be a mistake to dismiss the underlying concerns. Even skeptics of the most dire AI risk scenarios acknowledge clear evidence that AI tools are already transforming cybersecurity, with parallel developments in biorisk and alignment research. Restricting model releases alone is not a sufficient answer—it limits public access without addressing the root problems.
Ball’s analysis points to a more constructive path: trusting independent expert groups to guide the process, aligning behind the least-bad regulatory options rather than fighting every rule, and treating safety as an industry-wide concern rather than a competitive lever. That will be a difficult sell in an industry accustomed to moving fast and breaking things.
The stakes are high. If the approval process slows model development, it could put a chill on the massive data center buildout that underpins the entire AI sector. The next few weeks will test whether the industry can shift from internal rivalries to collective action—or whether the regulatory fog will simply deepen.
Frequently Asked Questions
Which AI models have been affected by government review?
Anthropic’s Fable and Mythos models were pulled from general release, and OpenAI’s GPT-5.6 is reportedly only available in a limited preview with customer-by-customer approval.
Why is the government reviewing AI models before release?
The reviews are intended to assess potential risks related to cybersecurity, biorisk, and AI alignment, though the specific safety standards have not been clearly defined by regulators.
What does this mean for the future of AI development?
The approval process could slow the pace of model releases, limit economic returns for costly AI systems, and potentially chill investment in data center infrastructure if delays become prolonged.
How are AI companies responding to the regulatory environment?
While some accuse Anthropic of regulatory capture and others accuse OpenAI of political maneuvering, the article argues that both labs now share the same fundamental problem: a need for a coherent, transparent release process.
Be the first to comment