🎧 Prefer to listen?

I was scrolling through AI news last week and realized something unsettling: the two biggest AI companies in the world can’t release their own models anymore without government sign-off. Not in the EU. Not in China. In the United States. If you’ve been following the AI landscape this year, you know tensions have been building — but this is a different kind of problem. It’s not about which company is winning. It’s about whether any of them can ship.

What actually happened

Here’s the sequence. Two weeks ago, the U.S. government pulled Anthropic’s Fable and Mythos models from public availability. The official reasoning centered on safety concerns, but the details were vague. Then The Information reported that OpenAI’s upcoming GPT 5.6 would face the same treatment: limited preview only, with the government approving access “customer by customer” before a general release could happen.

Anthropic’s Mythos has been stuck in preview for months with no timeline for general release. If GPT 5.6 gets trapped in a similar holding pattern, even a few weeks of delay could gut the economic return on what were enormously expensive models to train. And if the pace of model development slows, the data center buildout that’s been driving billions in investment starts looking a lot riskier.

The part that surprised me wasn’t that the government got involved. It’s that both companies — who’ve been bitter rivals — are now stuck in the exact same mess. The same process. The same uncertainty. The same existential risk to their business models.

Why this matters if you use AI tools (not just build them)

If you’re a solo creator, a small business owner, or someone who’s built workflows around AI tools, this might feel abstract. It’s not. Here’s why:

Feature releases slow down. When every new model needs government approval, the gap between “what’s possible” and “what you can actually use” widens. The tools you see in tutorials and demos may be months ahead of what’s publicly available.

Prices go up. Longer approval processes mean higher costs for AI companies. Those costs get passed to users. If you’re already paying for multiple AI subscriptions, expect pressure on pricing.

Competition stalls. One of the best things about the current AI market is that new players keep emerging — Lovable, Cursor, Claude’s new features. If releasing a frontier model requires months of government review, startups can’t afford to compete. The incumbents get entrenched by default.

Your automations break. If you’ve built workflows that depend on specific model capabilities, a delayed release means your next upgrade is on hold too.

The real problem: nobody knows the rules

What makes this situation particularly messy is that there’s no clear standard. As Dean Ball (a GMU fellow, now joining OpenAI) detailed in a recent analysis, the U.S. government doesn’t have the expertise or capacity for the kind of testing that would be needed here. It’s not even clear what regulators are trying to protect against — there’s been no public articulation of what specific risks triggered the Anthropic pull.

Compare this to how the government handles other consumer products. Pharmaceuticals go through FDA approval, but there’s a defined process with clear endpoints. Cars go through NHTSA testing with published standards. AI models are being held to… something. Nobody can say exactly what.

This ambiguity is the real threat. If you don’t know what “safe enough” means, every release becomes a political negotiation. And political negotiations don’t follow predictable timelines.

What the industry should be doing (but isn’t)

The smart take, as Ball outlined, is that AI companies need to stop treating regulation as a competitive weapon and start treating it as a collective problem. That means:

  • Trusting independent groups to guide the safety process, even when those groups don’t perfectly align with any one company’s interests
  • Lining up behind the least-bad regulatory options instead of fighting every proposal tooth and nail
  • Fighting for AI as an industry instead of seeing safety concerns as opportunities to ice out competitors

The problem is that the most prominent people in AI have billions riding on one company or another. Anthropic’s supporters see OpenAI’s government relationship as cozy corruption. OpenAI’s camp sees Anthropic’s safety advocacy as regulatory capture. Both narratives have some truth. Neither helps.

What I’m watching next

A few things will tell us whether this is a speed bump or a structural shift:

GPT 5.6’s actual release timeline. If the “couple of weeks” Altman reportedly projected holds, this is manageable. If it stretches to months like Mythos, the industry has a real problem.

Whether other countries follow. If the EU or UK adopt similar pre-release approval processes, the global AI market fragments overnight. What that means for solo creators could be significant.

Startup response. If new model releases slow, do startups pivot to building on existing models instead of training new ones? That would reshape the competitive landscape in ways that benefit no-code users — more tools built on stable foundations, fewer breakneck model upgrades.

The bottom line

The AI industry’s biggest threat right now isn’t a technical limitation — it’s a governance vacuum. When the government can pull your model without explaining why, and keep your competitor’s model in limbo indefinitely, the entire incentive structure for building frontier AI shifts. For those of us who use these tools daily, the practical impact is simple: expect slower feature rollouts, higher prices, and less competition in the months ahead. Stay flexible, build on stable APIs, and don’t assume the next model upgrade is coming on schedule.

If you want to build AI skills that work regardless of which models are available, start with the fundamentals at /start-here/.

Based on reporting by Russell Brandom at TechCrunch and analysis by Dean Ball at Hyperdimensional.