🤚 The Open-Palm Illumination

In the span of three weeks in June 2026, the United States government did something that no regulatory framework, no congressional hearing, and no think-tank white paper had ever managed to do: it pulled two of the most powerful AI models on Earth offline and told the companies that built them to sit down.

First came Anthropic’s Fable 5 and Mythos 5, yanked from global access on June 12 — just three days after launch — via an export-control directive that prohibited access for any foreign national, including Anthropic’s own employees. The stated reason? A jailbreak. The kind that, as several security researchers politely noted, every other frontier model can already do. Then OpenAI’s GPT-5.6 was confined to “a small group of trusted partners” whose identities were shared with the government, turning what should have been a standard product launch into something resembling a classified briefing with better catering.

The mechanism behind all of this? A June 2 executive order titled “Promoting Advanced Artificial Intelligence Innovation and Security,” which invited AI companies to voluntarily submit their frontier models for government review up to 30 days before release. The word “voluntarily” did a lot of heavy lifting in that sentence. When Anthropic launched Fable 5 without a government pre-brief, the White House demonstrated exactly how voluntary “voluntary” is when you have export-control authority and a phone.

👐 The Two-Handed Reality Check

Here’s the part that keeps constitutional lawyers awake at night and AI executives reaching for their anxiety medication: no specific criteria have been published for what triggers a restriction. No benchmarks. No thresholds. No defined list of which agencies have authority to act, or how long a review can last. The executive order established the idea of government oversight without establishing the rules of government oversight, which is a bit like installing a speed limit sign that just says “don’t.”

The NSA and CISA were directed to develop a classified benchmarking process to identify models with “advanced cyber capabilities.” Classified. As in, the companies being evaluated may not get to see the rubric by which they are being graded. Critics have called this a “de facto involuntary licensing regime” — a phrase that sounds like it was drafted by a lawyer who specializes in understatement.

Anthropic, to its credit, pushed back publicly. The company argued that the jailbreak cited as the trigger for Fable 5’s suspension was narrow, non-universal, and exposed only already-known, minor vulnerabilities. In other words: you just grounded our best model over something that your own government’s red-team tools could reproduce on any competing product. The directive was fully lifted on June 30, and Fable 5 returned to worldwide availability on July 1. OpenAI, meanwhile, took a more diplomatic approach, complying with the restrictions while stating that “we don’t believe this kind of government access process should become the long-term default.” Which is corporate for “please stop doing this.”

The Department of Commerce’s Center for AI Standards and Innovation conducted additional testing on GPT-5.6, with OpenAI sending technical experts to Washington to address concerns. This expedited review eventually cleared GPT-5.6 for its July 9 general availability launch — but not before the entire industry got a masterclass in what happens when your product roadmap requires a security clearance.

🌿 The Gentle Awakening

Step back far enough and the pattern becomes unmistakable. The United States government has discovered that it possesses an extraordinary power: the ability to ground a globally deployed AI model in a matter of hours, using export-control mechanisms that were designed for weapons technology and dual-use goods. No new legislation was required. No vote was held. The existing regulatory architecture was simply reinterpreted to cover software that thinks.

This is, depending on your perspective, either a necessary guardrail for technology that could reshape national security — or the most significant expansion of executive authority over commercial technology since the Atomic Energy Act. Possibly both. The uncomfortable truth is that frontier AI models now occupy the same regulatory grey zone as centrifuge components and satellite guidance systems: products that are legal to build, legal to sell, and subject to having their distribution halted by a phone call from a three-letter agency.

What makes this particularly fascinating is the speed asymmetry. AI companies can develop and deploy a model in weeks. The government’s review framework has no defined timeline. The June executive order’s 30-day pre-release window is voluntary, and the enforcement mechanism for non-compliance appears to be “we’ll use export controls instead.” The industry is moving at startup velocity while being regulated at bureaucratic velocity, and the friction between those two speeds is generating enough heat to power a data center.

👑 The Crown Verdict

We are witnessing the birth of a new regulatory paradigm, and like most births, it is messy, loud, and nobody in the room seems entirely sure what to do next. The White House has demonstrated that it can block frontier AI models. What it has not yet demonstrated is that it has a coherent framework for deciding when it should. Classified benchmarks, undefined triggers, and case-by-case enforcement create an environment where compliance is less about meeting standards and more about maintaining relationships — which is to say, it’s Washington working exactly as designed.

The real question isn’t whether the government should have oversight of frontier AI. It’s whether oversight conducted in darkness, without published criteria, subject to the whims of whoever occupies the Oval Office, constitutes oversight at all — or simply control by another name. As Peter Diamandis observed: “That’s an incredible power to give to the White House.” Indeed. And like most incredible powers, its value depends entirely on who’s wielding it and whether anyone thought to write an instruction manual.

Inspired by The Government Has Blocked AI | MOONSHOTS by Peter Diamandis.

Your regulatory exposure is showing. Comply wisely.