The Information Machine

US Government Export Control Directive Suspends Fable 5 and Mythos 5 for Foreign Nationals · history

Version 17

2026-06-22 18:33 UTC · 328 items

What

Ten days after the US Commerce Department directed Anthropic to suspend Fable 5 and Mythos 5 for foreign nationals, both models remain offline [20][21]. The BIS export licensing requirement from Commerce Secretary Lutnick [5] and the Pentagon's 'supply chain risk' designation [15] remain formally in place, while President Trump's June 20 statement that he no longer views Anthropic as a national security threat [22] has produced no policy reversal. A Financial Times analysis published June 22 argues Anthropic's own safety rhetoric — measured at 8x the volume of OpenAI's — may have contributed to triggering the ban [13], and Chinese open-weight model GLM-5.2 has emerged as a credible coding-agent alternative during the outage, adding competitive pressure on Anthropic's core revenue base [14].

Why it matters

The formal legal posture is unchanged while the economic consequences compound on two fronts: companies that migrated to non-US AI providers or open-weight models during the outage are making structural decisions that are difficult to reverse, and GLM-5.2's arrival offers the first open-weight model with credible coding-agent performance precisely when Anthropic's flagship models are offline. The FT's self-inflicted framing adds a dimension that complicates Anthropic's reinstatement negotiations.

Open questions

  • Trump stated on June 20 he no longer views Anthropic as a national security threat [22] — does this translate into formal action lifting the BIS licensing requirement and the Pentagon supply chain risk designation [15], or does it remain a public signal without policy follow-through?

  • The FT analysis argues Anthropic's safety rhetoric volume was self-defeating [13] — does Anthropic treat this as a communications lesson that shapes how it engages in reinstatement talks, or does it contest the framing?

  • GLM-5.2 is the first open-weight model Nathan Lambert considers a credible general coding agent, now competitive with Opus 4.5-class models [14] — does its emergence during the outage materially accelerate enterprise migration away from Anthropic, making commercial damage harder to reverse even if reinstatement occurs?

  • A single-sourced report suggests a reinstatement path via benchmark failure-rate tests rather than jailbreak elimination [23] — has this been corroborated, and does it represent a formal administration position or informal signaling?

Narrative

On June 12, 2026, the US Commerce Department issued a directive requiring Anthropic to suspend Fable 5 and Mythos 5 for all foreign nationals, citing a jailbreak that bypassed classifier-based safeguards for cybersecurity, chemistry, and biology prompts [1]. Amazon researchers demonstrated the jailbreak, and Amazon CEO Andy Jassy personally briefed senior Trump administration officials in conversations that directly preceded the government action — against a company in which Amazon holds a major investment [2][3]. Unable to restrict access by nationality on short notice, Anthropic disabled both models globally on June 13 [1]. Commerce Secretary Lutnick told Amodei, when Amodei said the directive meant the models must go offline: 'That's the point' [4]. On June 17, Lutnick formally sent Amodei a BIS letter placing both models under export licensing requirements [5].

The technical basis for the directive is contested. Katie Moussouris, the only outside expert given access to the government's report, assessed the triggering scenario — researchers used engineered code with deliberately planted fake vulnerabilities to extract patch-testing scripts through a multistep manual process — as 'the model working as intended' for cyberdefense [6][4]. Luta Security published a formal argument that the controls harm US cyber defense [7]. Zvi Mowshowitz argues the stated reinstatement path of fixing the jailbreak is technically impossible, since vulnerability identification and secure coding cannot be separated [8], and research confirms perfect jailbreak immunity is not achievable for any LLM [9]. An administration source framed reinstatement as requiring an 'attitude adjustment' rather than a specific technical fix [10], multiple sources confirm political hostility toward Anthropic's leadership was a factor [4][11], and Milk Road AI attributes the conflict to Amodei's consistent refusal to allow Claude for lethal applications [12].

A Financial Times analysis published June 22 adds a distinct framing: Anthropic's own sustained safety rhetoric may have contributed to bringing government scrutiny on itself. The FT measured five risk- or restriction-related words per 1,000 in Anthropic's 2026 publications, compared to 0.6 per 1,000 for OpenAI — an 8x disparity — and cited technologists who attribute the ban directly to Anthropic's repeated public warnings about AI's societal risks [13]. Separately, Nathan Lambert's June 22 analysis of GLM-5.2 — a Chinese open-weight model now competitive with Opus 4.5-class models in coding agent workflows — argues that Anthropic's Claude Code revenue, which had driven record growth, faces its first credible open-weight competition precisely during the models' suspension [14]. Lambert warns that if open models face comparable government restrictions, the result would be dangerous concentration of AI capability in a small number of closed providers.

Ten days into the outage, the formal legal posture is unchanged. The Pentagon has labeled Anthropic a 'supply chain risk' effective immediately [15], adding a DoD procurement designation alongside the BIS controls on a distinct legal track. Washington talks concluded June 16 without controls lifted; the US refused G7 allies and UK AISI special access [16][17]. Companies have signed backup contracts with non-US AI providers and deployed open-weight models in response to demonstrated government political risk [4], international teams received no advance warning before access was cut [18], and the refund window for API usage credits has closed [19].

Timeline

  • 2026-06-12: Anthropic publicly releases Fable 5 and Mythos 5. [32][33]
  • 2026-06-12: Amazon researchers demonstrate a jailbreak on Fable 5; Amazon CEO Andy Jassy briefs senior Trump administration officials whose conversations directly preceded the government action. [2][3][26]
  • 2026-06-12: White House issues Anthropic a 90-minute takedown ultimatum with no stated details on the actual threat. [11]
  • 2026-06-12: US Commerce Department issues an export control directive requiring Anthropic to suspend Fable 5 and Mythos 5 for all foreign nationals, citing the classifier-based jailbreak. [24][25][1]
  • 2026-06-12: Commerce Secretary Lutnick tells Anthropic CEO Amodei 'That's the point' when Amodei says the directive means the models must go offline. [4]
  • 2026-06-13: Anthropic disables Fable 5 and Mythos 5 globally after determining nationality-based restriction was not achievable on short notice. [1]
  • 2026-06-14: Nathan Lambert argues the episode opens a governance era of technically shallow, politically motivated AI restrictions and warns open-source advocates are unprepared for similar controls within two years. [28]
  • 2026-06-15: Semafor reports a China-linked group had accessed Mythos before the directive, raising model distillation concerns as a distinct driver of the government's action. [30]
  • 2026-06-15: Anthropic meets with the Commerce Department; an administration source says reinstatement requires an 'attitude adjustment' rather than a specific technical fix. [10]
  • 2026-06-16: Katie Moussouris assesses the triggering behavior as 'the model working as intended' for cyberdefense; Luta Security publishes a formal argument that the controls harm US cyber defense. [6][7]
  • 2026-06-16: Anthropic's Washington talks conclude without controls lifted; US refuses G7 allies and UK AISI special access. [16][17]
  • 2026-06-17: Commerce Secretary Lutnick sends a formal BIS letter to Anthropic CEO Amodei requiring a license before international export. [5]
  • 2026-06-17: Alex Stamos reports companies are signing backup contracts with non-US AI providers and deploying open-weight models in response to demonstrated US government political risk. [4]
  • 2026-06-18: Zvi Mowshowitz argues the government's stated reinstatement path — fixing the jailbreak — is technically impossible, since vulnerability identification and secure coding are inseparable. [8]
  • 2026-06-19: Research confirms perfect jailbreak immunity is not achievable for any LLM, deepening expert consensus that the government's stated fix path has no solution. [9]
  • 2026-06-20: In an Axios interview, President Trump states he no longer views Anthropic as a national security threat, while acknowledging he had held that view 'a week ago.' [22][12]
  • 2026-06-21: Pentagon labels Anthropic a 'supply chain risk' effective immediately, adding a DoD designation alongside the BIS export licensing controls. [15]
  • 2026-06-21: Fable 5 and Mythos 5 remain offline on Day 9; the refund window for API usage credits has closed and international teams received no advance warning when access was cut. [20][21][19][18]
  • 2026-06-22: Financial Times analysis argues Anthropic's safety rhetoric — 5 per 1,000 words vs. 0.6 for OpenAI — may have contributed to triggering the export ban. [13]
  • 2026-06-22: GLM-5.2, a Chinese open-weight model, reaches Opus 4.5-class capability in coding agent workflows, offering the first credible open alternative to Claude Code during the Anthropic outage. [14]

Perspectives

Anthropic

Complying under legal obligation while contesting the directive as technically disproportionate; Washington talks concluded June 16 without reinstatement and the formal BIS letter remains in place.

Evolution: Consistent on technical contest; any reinstatement path involves political accommodation. A single-sourced report of a benchmark-based reinstatement path remains unconfirmed.

[24][1][10][16][4]

Trump / White House

Trump stated on June 20 he no longer views Anthropic as a national security threat; an administration source separately framed reinstatement as requiring an 'attitude adjustment.'

Evolution: Publicly softened from the June 12 national security threat framing, but Trump's statement diverges from Lutnick's formal BIS action and the Pentagon's supply chain designation, with no announced policy change.

[10][22][12]

Commerce Department / Secretary Lutnick

Escalated from informal directive to formal BIS licensing requirement; 'that's the point' exchange confirms model suspension was the intended outcome; refused G7 allies and UK AISI access.

Evolution: Consistent on enforcement; formal actions remain in place even as Trump's public statements have softened.

[24][25][16][17][5][4]

Pentagon / Department of Defense

Labeled Anthropic a 'supply chain risk' effective immediately, adding a DoD procurement designation alongside the BIS export licensing controls on a distinct legal track.

Evolution: Entered as a new institutional actor on June 21; acts independently of the Commerce BIS controls.

[15]

Amazon / Andy Jassy

Amazon researchers identified and demonstrated the jailbreak; Jassy personally briefed administration officials in conversations that directly preceded the government action against a company in which Amazon holds a major investment.

Evolution: Consistent; proximate causal role confirmed.

[2][3][26]

Luta Security / Katie Moussouris

The triggering scenario used deliberately planted fake CVEs on engineered code and is 'the model working as intended' for cyberdefense; published a formal argument that the controls harm US cyber defense.

Evolution: Consistent.

[6][7][4]

Zvi Mowshowitz

Directive is technically incoherent and politically driven; the stated reinstatement path is technically impossible since vulnerability identification and secure coding are inseparable.

Evolution: Consistent; June 18–19 work sharpened the technical impossibility argument without changing the core critique.

[27][11][4][8][9]

Nathan Lambert / Financial Times / technology and national security observers

Lambert argues the episode opens a governance era of politically motivated AI restrictions and warns GLM-5.2's arrival creates real competitive pressure on Anthropic's Claude Code revenue during the outage; the FT argues Anthropic's own 8x-higher safety rhetoric volume contributed to triggering the ban.

Evolution: Lambert's analysis deepened from a governance-era warning to include specific competitive economic analysis; the FT's self-inflicted framing is new and cuts against Anthropic's posture of disproportionate government overreach.

[28][4][29][14][13]

Tensions

  • Anthropic, Moussouris, and Luta Security assess the triggering behavior as standard defensive security work on engineered fake code; the government treated it as justifying full model suspension and formal BIS licensing without publicly addressing the expert counter-assessment. [6][7][4][24][5]
  • Trump publicly stated he no longer views Anthropic as a national security threat; Lutnick's formal BIS letter of June 17 remains in force and the Pentagon has separately labeled Anthropic a supply chain risk, leaving Trump's public softening and the formal legal posture in divergence. [22][5][12][15]
  • The government's stated reinstatement path requires fixing the jailbreak; Zvi argues this is technically impossible and research confirms perfect jailbreak immunity is unachievable; Lutnick's 'that's the point' exchange and the 'attitude adjustment' framing indicate the actual standard is political accommodation, not a technical bar. [27][4][10][5][8][9]
  • The FT attributes the export ban partly to Anthropic's own safety rhetoric — quantitatively 8x heavier than OpenAI's — framing Anthropic as having contributed to its own predicament; Anthropic and Moussouris frame the directive as disproportionate government overreach with no legitimate technical basis. [13][6][7][24]
  • Zvi and multiple sources argue political hostility toward Anthropic's leadership drove the action; Milk Road AI attributes it to Amodei's refusal on lethal use; the government's public framing emphasizes the jailbreak and Chinese model distillation risk. [4][11][12][24][30]
  • Lambert predicts open-source AI models will face comparable government restrictions within two years and warns this would dangerously concentrate AI power; Just Security argues such extensions would not advance US competitive interests given cheaper Chinese alternatives already in enterprise use. [28][31][14]

Sources

  1. [1] Anthropic shuts down Fable, Mythos models following Trump admin directive — Ars Technica AI (2026-06-13)
  2. [2] Reuters: Amazon’s Andy Jassy was among the people who warned senior Trump officials this week about security concerns ar… — Rohan Paul Twitter (2026-06-13)
  3. [3] Amazon CEO’s Talks With U.S. Officials Triggered Crackdown on Anthropic Models - WSJ — reactive:fable-mythos-export-control
  4. [4] The Once And Future Fable #3: Fix This Code — Zvi's AI Roundups (2026-06-17)
  5. [5] Full Letter From Commerce Secretary Howard Lutnick to Dario Amodei — Rohan Paul Twitter (2026-06-17)
  6. [6] Quoting Matteo Wong, The Atlantic — Simon Willison (2026-06-16)
  7. [7] The Fable 5 Export Controls Harm US Cyber Defense - Luta Security — reactive:fable-mythos-export-control
  8. [8] AI #173: AI Pauses — Zvi's AI Roundups (2026-06-18)
  9. [9] Perfect immunity from jailbreak is not possible even for the strongest of LLMs. — Rohan Paul Twitter (2026-06-19)
  10. [10] "They screwed us": Personality clashes sent Anthropic's models offline — Simon Willison (2026-06-15)
  11. [11] The Once And Future Fable #2 — Zvi's AI Roundups (2026-06-15)
  12. [12] Trump just told the world that a week ago he considered Anthropic a national security threat (Save this). — Milk Road AI Twitter (2026-06-19)
  13. [13] How Anthropic may have talked itself into an AI export ban — Ars Technica AI (2026-06-22)
  14. [14] GLM-5.2 is the step change for open agents — Interconnects (2026-06-22)
  15. [15] Pentagon says it is labeling AI company SF-based Anthropic a supply chain risk 'effective immediately' - ABC7 San Francisco — reactive:fable-mythos-export-control
  16. [16] Anthropic Washington Talks End Without Lifting Export Controls — reactive:fable-mythos-export-control (2026-06-16)
  17. [17] The US just refused to give G7 allies special access to Anthropic’s Mythos 5 and Fable 5. — Rohan Paul Twitter (2026-06-16)
  18. [18] RT @PsudoMike: US export control order pulled Fable 5 and Mythos 5 from Canadian teams on hours notice. No warning. No m... — reactive:fable-mythos-export-control (2026-06-20)
  19. [19] Claude Fable 5 + Mythos 5 are STILL offline. Day 9 of the export control ban. The refund window for usage credits closed... — reactive:fable-mythos-export-control (2026-06-21)
  20. [20] JUST IN: Anthropic's Fable 5 and Mythos 5 are still down globally after a US export-control directive — reactive:fable-mythos-export-control (2026-06-21)
  21. [21] 🚨 Day 9 of the CLAUDE FABLE 5 BAN! — reactive:fable-mythos-export-control (2026-06-21)
  22. [22] Not anymore: Trump on whether he sees Anthropic threatening national security — Rohan Paul Twitter (2026-06-20)
  23. [23] The White House and Anthropic may have found the first serious path to restore Mythos and Fable access without pretendin… — Rohan Paul Twitter (2026-06-19)
  24. [24] Statement on the US government directive to suspend access to Fable 5 and Mythos 5 — Anthropic News (2026-06-12)
  25. [25] Scoop: Trump admin blocks foreign access to Anthropic's most powerful AI — reactive:fable-mythos-export-control
  26. [26] Amazon CEO reportedly raised Anthropic model concerns before ... — reactive:fable-mythos-export-control
  27. [27] American Government Takes Down Claude Fable — Zvi's AI Roundups (2026-06-13)
  28. [28] Welcome to the AGI era of AI governance — Interconnects (2026-06-14)
  29. [29] Technology, National Security Leaders Say Anthropic and Trump’s ‘Messy Public Breakup’ Will Harm Both — reactive:fable-mythos-export-control
  30. [30] A new Semafor report says the White House partly decided to place export restrictions on Anthropic’s Mythos over concern… — Rohan Paul Twitter (2026-06-15)
  31. [31] Export Controls on Open-Source Models Will Not Win the AI Race — reactive:fable-mythos-export-control
  32. [32] Anthropic releases Claude Fable 5, its most powerful public AI model — reactive:fable-mythos-export-control
  33. [33] Introducing Claude Fable 5 and Claude Mythos 5 - Claude API Docs — reactive:claude-fable-5-mythos-launch