CNBC reported on July 17 that the White House has started dictating which partners get pre-release access to frontier AI models from OpenAI, Anthropic and Google, a power that used to sit entirely with the labs.
Until this year, Dario Amodei and Sam Altman decided who saw their most powerful models before the public did. Not anymore. Now, according to CNBC, the Trump administration is inserting itself directly into that decision, telling companies which "trusted partners" can get early access and which cannot. It's a quiet but real transfer of power. It traces back to an executive order Trump signed on June 2 that asked AI developers to hand federal agencies up to 30 days of pre-release access to any model the government designates as a "covered frontier model." The order also directs agencies to help select which outside partners get first access. Voluntary, on paper. In practice, the labs are complying.
You don't have to guess how this plays out. Anthropic already lived it. The company launched Claude Fable 5 and Mythos 5 on June 9. Three days later, the Commerce Department ordered a worldwide suspension of both models for foreign nationals, citing export control authority and national security. Commerce Secretary Howard Lutnick sent the order directly to Amodei. The trigger, according to Forbes and Fortune, was a jailbreak that Amazon researchers found: prompted the right way, Fable 5 would generate exploit code for a real software vulnerability. White House AI adviser David Sacks said Anthropic refused to fix it. Anthropic disputed how severe the flaw actually was and criticized the process as opaque. The standoff lasted 19 days. Commerce lifted the export controls on June 30, and Anthropic began restoring global access to Fable 5 on July 1.
OpenAI hit a version of the same wall. The administration told the company to gate its GPT-5.6 cybersecurity model behind a consortium it calls Daybreak, limiting who can touch it before wider release. Google, meanwhile, walked through untouched. Its Gemini 3.5 Pro launched without restrictions because it didn't cross the capability threshold the government uses to flag a model as covered. Same regulatory regime, three completely different outcomes, decided model by model rather than by any fixed rule.
If your startup's roadmap depends on API access to whatever OpenAI or Anthropic ships next, you no longer just need to watch the lab's release notes. You need to watch Washington's calendar too. A 30-day government review window, plus however long a dispute like Anthropic's takes to resolve, is real time you might not get back. Nineteen days doesn't sound like much until it's the nineteen days your product launch was supposed to happen in.
And the bottleneck cuts both ways. While Fable 5 sat offline, Chinese open-weight labs kept shipping. CNBC reported on June 30 that the crackdown on US frontier models has handed an opening to companies like Moonshot AI and Z.AI, whose Kimi and GLM model families are catching up fast on cost and, increasingly, on capability. Coinbase CEO Brian Armstrong has said the company now runs GLM-5.2 and Kimi in production, cutting its AI spend nearly in half even as usage climbed. Travis Lanham, co-founder of the security startup Armadin, has pointed to Moonshot's models showing real gains on cybersecurity tasks specifically, the same category the White House is busiest gatekeeping at home.
A White House spokesperson pushed back on the framing.
Also read: OpenAI's GPT-5.6 Sol can already find browser exploits, just not finish them • SpaceX is reportedly chasing a Pentagon AI compute deal • AWS billing dashboard told a customer who spent 19 cents that they owed 2.5 billion dollars
