Pulse

Legal risk / Jun 29, 2026 / 5 min

Commerce Missed Congress's Deadline on the Fable 5 Order

On June 26, bipartisan House members' deadline for Commerce to explain the legal basis for shutting down Anthropic's Fable 5 expired — and as of June 29, Secretary Lutnick still has not published an answer while GPT-5.6 inherits the same precedent.

Thesis Washington's frontier-AI kill switch runs on export law nobody can cite — and Commerce's silence past Congress's deadline proves the regime governing America's most valuable private companies is improvised, not lawful.

Congress gave Commerce until June 26 to explain how export law became a kill switch on live AI models — and as of June 29, Secretary Howard Lutnick still hasn't published a legal answer, even as he partially restored Mythos 5 for a federal guest list while Fable 5 stays dark.

What's new: On June 18, four House members — Sam Liccardo (D-CA), Jay Obernolte (R-CA), Ted Lieu (D-CA), and Scott Franklin (R-FL) — sent Lutnick a bipartisan letter demanding the statutory authority behind the June 12 directive that knocked Anthropic's Fable 5 and Mythos 5 offline worldwide, per POLITICO and CryptoBriefing.

Their June 26 deadline passed with no public Commerce response, per tracking reports as of June 29. The Bureau of Industry and Security still shows no published written justification for treating commercial API access as an export.

Why it matters: The non-answer is not a footnote — it is the foundation Washington is building on.

  • On June 12, Lutnick ordered Anthropic to block foreign nationals from accessing Fable 5 and Mythos 5 under the Export Administration Regulations. Anthropic disabled both models for everyone because it could not reliably screen users by nationality in real time, per CryptoBriefing.
  • On June 26, Lutnick wrote Anthropic a different letter — partially restoring Mythos 5 for roughly 100 entities on an Annex A allow list. Fable 5 restrictions unchanged. Axios reported Lutnick reserved the right to re-evaluate and adjust license requirements at any time.
  • Congress's questions about legal authority, evidentiary thresholds, and process? Still unanswered in public.

The legal fight: Export law governs the "export of items" — historically semiconductors and hardware crossing borders, not a live API call from a domestic developer.

  • House lawmakers specifically asked what statute Lutnick invoked to restrict foreign nationals from accessing an AI API rather than physical goods or exportable code, per AI Weekly's summary of Washington Post reporting.
  • Bloomberg reported Lutnick's order marked an "unprecedented use of export control laws" targeting mere usage of cutting-edge AI models.
  • Legal analysts quoted in coverage argue EAR § 744.22 was not written for API inference — and there may be no "export" under existing definitions when weights never leave the server.

Who asked: The letter's signatories span both parties and the AI policy spectrum — not a regulation fight, but an authority fight.

Liccardo and Lieu are Democrats. Obernolte co-chaired the bipartisan House AI Task Force and holds a Stanford PhD in AI. Franklin is a Florida Republican on the Science Committee.

Their letter, per POLITICO: "While this action concerned a single AI model, it appears to represent a significant new application of export control authorities to advanced AI systems and therefore raises important questions for the broader U.S. AI ecosystem."

They demanded to know:

  • What analysis Commerce used
  • What standards and criteria triggered the controls
  • Whether the capabilities are unique to Anthropic
  • How those standards apply going forward

The evidence gap: The administration's stated trigger, per Axios: Amazon raised jailbreaking concerns about Mythos, and Anthropic failed to respond seriously. The identity of the jailbreak, the nature of the vulnerability demonstrated, and independent verification have not been publicly released.

That opacity matters because the jailbreak claim is the entire factual predicate for shutting down a $1 trillion company's flagship products.

What changed this weekend: GPT-5.6 Sol launched under the same logic days later.

  • Washington limited OpenAI's debut to roughly 20 government-vetted partners with customer-by-customer approval, extending the precedent without answering whether Commerce even has the power to impose it.
  • If API inference counts as export, every frontier lab's hosted model is one Lutnick letter away from a global shutdown.

Convina's view: You cannot govern a trillion-dollar industry through secret letters citing trade law written for silicon wafers. Commerce's silence past Congress's deadline is not administrative delay — it is an admission that the legal theory cannot survive daylight. Washington improvised a kill switch, partially reopened one model for a guest list, and left the statutory question unanswered. That is not security policy. It is sovereign risk with no disclosure requirements — and every enterprise buyer signing an API contract this summer is underwriting it.

Research Signals

https://subscriber.politicopro.com/article/2026/06/lawmakers-want-answers-from-commerce-on-anthropic-export-controls-00968444 https://cryptobriefing.com/lawmakers-demand-answers-anthropic-ai-restrictions/ https://www.axios.com/2026/06/27/commerce-anthropic-mythos-restrictions-lift https://www.bloomberg.com/news/articles/2026-06-19/lutnick-s-anthropic-crackdown-claims-new-power-over-ai-models https://aiweekly.co/alerts/house-lawmakers-press-lutnick-on-legal-basis-for-anthropic-ai-ban