The era of the AI arms race, defined by a frantic sprint between OpenAI and Anthropic, has effectively ended. It has been replaced by a new, more rigid reality: the U.S. government is now the final arbiter of what gets released, and it is moving with the speed of a bureaucracy, not a startup.

Two weeks after regulators effectively mothballed Anthropic’s Fable and Mythos models, OpenAI is facing the same wall. Reports indicate that the company’s latest model, GPT-5.6, will be restricted to a limited preview, with the government approving access on a customer-by-customer basis. While Sam Altman has reportedly projected this limbo might last only a few weeks, the reality of the current regulatory environment suggests otherwise. Anthropic’s Mythos has been trapped in a similar preview state for months with no clear path to general release.

This isn't just a corporate headache; it is a structural shift for the entire sector. For the labs, every week spent in a government-mandated holding pattern is a direct hit to the bottom line. For the broader tech ecosystem, it signals a potential chill on the massive data center buildouts that have fueled the AI boom. If the models can’t reach the public, the infrastructure supporting them becomes a liability rather than an asset.

The Myth of Competitive Advantage

Conversations in Silicon Valley remain obsessed with the narrative of rivalry. Industry insiders are busy trading accusations: some claim Anthropic is playing a game of regulatory capture to slow down its larger competitors, while others argue OpenAI is cozying up to the Trump administration to ice out smaller players.

These theories are seductive because they fit the old playbook of tech competition. But they miss the scale of the current problem. The government’s current approach to model approval is haphazard and lacks a clear framework. There is no version of this regulatory process that helps one lab without creating a bottleneck for the others.

As Dean Ball, a fellow at the Mercatus Center, recently noted, the U.S. government currently lacks the technical capacity and the defined safety criteria to conduct the kind of testing it is demanding. Regulators are effectively asking for a safety check without being able to articulate exactly what they are trying to prevent. When the rules are opaque, the result is not safety—it is stagnation.

Why Real Risks Require Collective Action

It is easy to dismiss the government’s caution as bureaucratic overreach, but the underlying concerns are not imaginary. We are seeing AI tools fundamentally alter the landscape of cybersecurity, and the potential for misuse in areas like biorisk is no longer a theoretical exercise.

Restricting releases is a blunt instrument, but the industry has yet to offer a sharper one. If the labs continue to treat safety as a competitive wedge rather than a shared burden, they will lose the ability to shape the rules entirely.

Moving forward will require a fundamental change in strategy. It means:

  • Trusting independent oversight: The industry must accept third-party testing groups, even those that don't perfectly align with their internal goals.
  • Aligning on the 'least-bad' regulation: Instead of fighting every legislative attempt, labs need to coalesce around a unified, sensible framework that provides a clear path to market.
  • Ending the zero-sum game: The industry must stop viewing safety as an opportunity to gain an advantage over rivals.

Key Takeaways

  • The Gatekeeper Effect: The U.S. government is now effectively controlling the release cadence of frontier AI, creating a bottleneck that threatens the economic viability of new models.
  • Regulatory Uncertainty: There is currently no clear, standardized process for safety testing, leaving models like GPT-5.6 and Mythos in indefinite limbo.
  • A Call for Unity: The industry’s focus on internal rivalry is blinding it to the need for a collective, industry-wide strategy to engage with regulators before the current, haphazard process becomes permanent.

For the engineers and executives at the center of this, the transition from 'move fast and break things' to 'wait for approval' is a bitter pill. But the capabilities of these models have reached a point where they carry genuine political consequences. The industry’s ability to survive this transition will depend on whether it can act as a unified sector or if it will continue to fracture while the government holds the keys.