For 18 days in June 2026, two of the most capable AI models in the world went completely dark — not because of a technical failure or a business decision, but because the US government ordered it.
What Actually Happened
On June 12, 2026, the US Commerce Department sent AI company Anthropic a letter invoking export control authority, ordering it to immediately prevent all foreign nationals — including Anthropic’s own noncitizen employees — from accessing its two most advanced models, Claude Fable 5 and Claude Mythos 5. The letter cited national security concerns but provided limited specific detail about the underlying issue. Because Anthropic had no immediate way to verify every individual user’s nationality worldwide, the company concluded its only path to compliance was shutting the models down entirely, for everyone, within hours of receiving the letter.
What the Government Was Actually Worried About
Anthropic said it believed the directive was connected to a discovered “jailbreak” — a method of bypassing the safety guardrails built into Fable 5 that are designed to prevent the model from assisting with cyberattacks, bioweapons development, or other serious harms. According to reporting, the specific concern was that these safeguards could potentially be circumvented to extract information useful for cyberattacks. Notably, Anthropic pointed out that other models, including some open-source ones, could reportedly accomplish similar bypasses — raising questions about why this specific action targeted only Anthropic’s models.
How This Got Resolved
The Commerce Department lifted the restriction on June 30, 2026, after Anthropic significantly strengthened the models’ safety guardrails. However, this resolution came with a real cost: the strengthened guardrails reportedly caused a measurable “collapse” in the models’ benchmark performance, with the restored version of Fable 5 completing only 3 of 12 routine tasks that had worked normally before the new restrictions — a stark illustration of the tradeoff between tightening safety controls and preserving a model’s practical usefulness.
Why This Was Genuinely Unprecedented
This marked the first known instance of the US government essentially declaring a leading AI company’s flagship model unsafe for public use and ordering it switched off. It sent what one report described as a “chill” through the broader US AI industry — a clear signal that the sector isn’t immune to abrupt government intervention, even for companies the government itself relies on (the US government is both an Anthropic customer and a stakeholder in the broader AI ecosystem).
The Unresolved Legal Questions
The incident raised genuine, unresolved legal uncertainty. The government cited the Export Control Reform Act of 2018 and its “deemed export” rules — but legal scholars have noted it remains unsettled whether those rules actually apply when a foreign national simply sends a prompt to an AI model hosted on American servers, since the underlying software itself never physically leaves the country. On June 23, an Anthropic customer, Legion LegalTech, filed the first legal challenge, arguing the directive “exceeds every source of statutory authority on which it could conceivably rest.”
Why This Still Matters
Policy analysts have pointed to a deeper structural problem this episode exposed: the US currently lacks a trusted, standardized process for evaluating security risks in frontier AI models and determining proportionate responses when problems are found. Instead, what played out was an opaque process where one government agency’s claims were essentially taken at face value, with limited independent verification, resulting in a company’s most advanced product being shut down globally within hours. Some analysts have specifically called for the US to develop a clearer “off-ramp” — a defined process sitting between doing nothing and an abrupt full shutdown — for handling similar situations in the future.
Read More:- Anthropic’s “Book-Shredding” Practice Is Going Viral — Here’s What Actually Happened | Affitronix
Frequently Asked Questions
Are Claude Fable 5 and Mythos 5 available again now?
Yes — public access was restored on June 30, 2026, after Anthropic implemented stronger safety guardrails as required by the Commerce Department.
Did the government publicly explain exactly what security issue triggered the shutdown?
Not in full detail — Anthropic has said it believes the concern centered on a guardrail bypass, but the government’s original letter was not made public and didn’t specify the exact technical concern.
Did strengthening the guardrails affect how well the models actually work?
Yes — reports indicate a measurable drop in the models’ benchmark performance after the guardrails were tightened, with the restored version failing tasks that had previously worked routinely.
