Anthropic Disclosed. OpenAI Did Not. One Ran.

The June 12 Order

Fable 5 was suspended for a jailbreak, a technique that bypasses a model's output constraints by prompting it to read a codebase and surface software flaws. 5 carries the identical demonstrated jailbreak and continued taking production calls through the weekend. At 5:21pm ET on June 12, Anthropic received a directive ordering suspension of Fable 5 and Mythos 5, its two newest production models, for "any foreign national, whether inside or outside the United States," on national security grounds. No agency was named. Anthropic disabled both models globally within hours because it could not segregate foreign nationals from its user base in real time. Paying subscribers who had unlocked free access through June 22 lost it the next morning.

5 (OpenAI's current production model) as subject to the identical demonstrated jailbreak. 5. A team that built its production AI stack on Anthropic's API is now rebuilding. A team that built the same stack on OpenAI's is not. As of June 14, the Singapore bank's production calls go unanswered. The competitor model carrying the same jailbreak took calls through the weekend.

One Jailbreak, Two Outcomes

5 simultaneously. 5 equivalence in the same breath as the suspension. The directive pulled one. The operative criterion, by revealed behavior, was not the vulnerability. It was Anthropic's voluntary public disclosure of that equivalence. For any lab deciding what to disclose about its own jailbreak surface, that distinction is now the policy reference case.

The BIS, which administers US chip export controls, published guidance on June 1 closing a loophole that had let Chinese-affiliated companies in Singapore, Dubai, and other third-country hubs acquire Blackwell-tier Nvidia graphics processors (Nvidia's current flagship AI chip generation) by routing purchases through local subsidiaries. An importer that had been using that routing now needs a BIS export license to continue. That instrument governs compute supply. The June 12 directive governed deployed model access and arrived from a national security authority the directive did not name.

A developer in Seoul running production traffic on Anthropic's API learned June 12 that a US national security directive can disable their stack globally within hours. The BIS instrument governs compute supply and names an enforcing agency. The June 12 directive governed deployed model access and named none. OpenAI's stack ran through the weekend.