The Government Just Set a Powerful New AI Precedent

The Trump administration just showed it can turn off cutting-edge AI for the entire world with a single secret letter.

Story Snapshot

The US Commerce Department ordered Anthropic to shut down its Fable 5 and Mythos 5 AI models worldwide over national security fears.
Officials said a “jailbreak” could help find software flaws, but they have not released technical proof or a public incident report.
After two weeks, the government quietly granted limited access back to “trusted” US companies and agencies, saying safeguards are now enough.
This is the first known time US export-control powers were used to directly switch off a specific frontier AI model, setting a major precedent.

A sudden global shutdown shows how much power Washington now holds over AI

On June 12, 2026, the US Department of Commerce sent Anthropic a legally binding export-control directive that hit at 5:21 p.m. Eastern time. The order told the company to suspend access to its newest and most capable AI models, Fable 5 and Mythos 5, for any foreign national, whether they lived abroad or inside the United States. Because Anthropic said it cannot reliably sort users by nationality in real time, it shut both models off for everyone on Earth to avoid civil and criminal penalties.

The directive was issued through the Commerce Department’s Bureau of Industry and Security and signed by Commerce Secretary Howard Lutnick, using national security powers that usually apply to weapons or sensitive chips. Officials said they were worried about a “coding task jailbreak” that could bypass safety systems and help users find software vulnerabilities. That made Mythos 5, which was already known to be very strong at security testing, look like a tool that could be turned into an automated hacking engine if misused.

National security fears collide with a lack of public evidence

The government described the jailbreak as “catastrophic,” but has not released any detailed technical report, public incident, or written appendix to back that claim. Anthropic says it reviewed the demonstration and found that the exploit used narrow, already known weaknesses similar to ones seen in other frontier models, including OpenAI’s GPT‑5.5, and that it had already added defenses. Independent security voices cited by FifthRow called the issue “narrow” and “non‑universal,” raising doubts that this one flaw justified a total global shutdown.

The trigger for the directive reportedly came from outside government: Amazon security researchers found the jailbreak and shared a report with Anthropic. That corporate involvement worries people on both the left and right who already suspect big tech firms influence Washington to shape markets, not just protect the country. Critics say that if this level of fear over a single, disputed exploit became the standard for export controls, almost every powerful AI model would need to be frozen, turning the Commerce Department into a kind of global AI gatekeeper.

A partial rollback that still leaves many users shut out

Two weeks after the shutdown, Commerce Secretary Lutnick sent Anthropic a formal exemption letter dated June 26, 2026. The letter allowed Mythos 5 to be released again, but only to about 100 “trusted” companies and US federal agencies, and only because the Secretary said “appropriate safeguards are in place.” This carve‑out shows how Washington now picks who may use the most advanced AI: not regular small businesses, not foreign researchers, but a narrow group of approved insiders.

Anthropic has said it is working with the government to restore broader access and that it sees the directive as based on a “misunderstanding” of the real risk. At the same time, some users, including a video host and people in the Philippines, reported that they still had Fable 5 access after the supposed global shutdown, hinting at uneven enforcement. That inconsistency fuels anger for many Americans who already believe federal rules hit honest users and small players harder than well‑connected firms who know how to navigate or bend the system.

A new AI export precedent deepens fears of an unaccountable “deep state”

Legal analysts say this is the first known use of US export-control powers to directly target a specific frontier AI model already deployed at scale, not just its training data or the chips that power it. It fits into a larger pattern: since 2024, the federal government has steadily expanded export rules to cover advanced model weights, data centers, and even “catch‑all” items that might aid foreign militaries or intelligence services, often based on broad risk claims.

Anthropic is brinding back Mythos/Fable access. With cybersecurity uses disabled/limited. They also confirm that GPT-5.5 nd Kimi K2.7 models were able to find the same security vulnerability which caused enacting export controls. The US Government will now get a pre‑release… pic.twitter.com/BTdpBJapyR

— Lukasz Olejnik (@lukOlejnik) July 1, 2026

For conservatives, this episode looks like more proof that global tech and security elites can shut down innovation and pick winners and losers, while everyday Americans get higher costs and fewer choices. For liberals, it highlights secretive national security moves that lack transparency, ignore civil fairness, and seem to serve corporate interests. Both sides see the same thing: a powerful federal system that can flip a switch and kill access to vital tools, without sharing the evidence or facing real public debate.