The directive is two days old. The legal fight is already underway.
Commerce Secretary Howard Lutnick issued an export control directive on June 12, 2026, ordering Anthropic to suspend global access to Claude Fable 5 and Claude Mythos 5 on national security grounds, according to Politico’s reporting on the 24-hour White House decision process that preceded it. Anthropic complied, both models went dark globally, but compliance didn’t mean agreement.
Anthropic’s response, published on the company’s website under the title “Where things stand with the Department of War,” makes the statutory argument explicit. The company characterizes the directive’s scope as narrow and points to 10 USC 3252 as the governing authority, a statute that exists, per Anthropic’s own framing, to protect a specific and limited category of defense-relevant technology. Anthropic’s position is that the vulnerability the directive was premised on falls outside what the statute was designed to address.
The jailbreak at the center of this is only partially verified. Multiple reports confirm a third-party demonstration of a technique bypassing Fable 5 and Mythos 5 safeguards prompted the directive. The specific technique name and the identity of the party who reported it couldn’t be independently verified and won’t be reported here as established fact. What Anthropic’s own statement acknowledges, per the cross-reference excerpt, is that the identified issue was narrow. That acknowledgment matters: it means the dispute isn’t about whether a vulnerability existed. It’s about whether that vulnerability warranted a full global suspension under 10 USC 3252.
Verification
Partial Anthropic public statement (vendor); Politico (T3 reporting) Jailbreak technique name and reporting party identity unverified. Epoch AI benchmark figure (88% FrontierMath T4 v2) pending URL resolution. SWE-Bench Pro 80.3% is vendor-reported only.Anthropic also stated, according to its public response, that other commercially available models possess comparable capabilities to identify the vulnerabilities in question. This is a vendor claim, not independently verified. But it’s strategically significant: if accurate, it undercuts the national security logic for singling out Fable 5 and Mythos 5.
The benchmark picture for Fable 5 is mixed on verification. Fable 5 reportedly scored 88% on FrontierMath Tier 4 (v2) per Epoch AI’s leaderboard, according to data cited at the time of its suspension, independent confirmation of that figure is pending source resolution. Anthropic’s internal evaluation puts SWE-Bench Pro at 80.3%, which is a vendor-reported figure only.
Here’s the catch: while the legal argument plays out, the operational disruption is live. Every enterprise that built workflows on Fable 5’s API, including its 1M-token context window, is running on a suspended model with no confirmed restoration timeline. The legal dispute doesn’t pause the outage.
Definition
Don’t expect a fast resolution. A statutory challenge to a national security directive moves through channels that don’t favor speed. In the meantime, compliance teams at enterprises with Anthropic-dependent deployments face a straightforward question: what’s the contingency?
The real question isn’t whether Anthropic’s statutory argument holds. It’s whether this directive establishes a precedent that any US-controlled frontier AI model can be suspended at Commerce’s discretion, and what that means for enterprise AI procurement strategy going forward. 10 USC 3252 may be narrow. The precedent being set is not.