# Anthropic Model Export Directive On June 12, 2026, the U.S. government issued an export control directive ordering [[Anthropic]] to immediately suspend all access to [[Claude Fable 5]] and [[Claude Mythos 5]] by any foreign national — whether inside or outside the United States, including foreign-national Anthropic employees. The directive, delivered at 5:21 PM ET on a Friday, cited national security authorities and forced Anthropic to abruptly disable both models for all customers to ensure compliance. Access to Anthropic's other models (Opus 4.8 and below) was not affected. This was the first AI-specific export control action in U.S. history — a government recall of a commercial AI model based on a discovered vulnerability, rather than a pre-deployment restriction. ## The Trigger: A Jailbreak The government's stated rationale was the discovery of a method to bypass [[Claude Fable 5]]'s safety guardrails. Fable 5 is the publicly available, safety-constrained version of the Mythos-class model. [[Claude Mythos 5]] — the same model with guardrails lifted — has extraordinary cybersecurity capabilities: it can find software vulnerabilities better than all but the most skilled humans and had already identified over 10,000 high- or critical-severity flaws through Project Glasswing. The government's concern: if Fable 5's guardrails could be bypassed, adversaries could effectively access Mythos-class vulnerability-detection capabilities through a publicly available model. The jailbreak turned a controlled-release safety model into a potential national security risk. ## Anthropic's Response Anthropic complied with the directive but publicly disagreed with its scope, arguing that "if this standard was applied across the industry, it would essentially halt all new model deployments for all frontier model providers." The company characterized the jailbreak as "narrow" and noted that vulnerability to prompt-injection and guardrail bypasses is an industry-wide challenge, not unique to Fable 5. The subtext: Anthropic was being punished for making Mythos-class capabilities available to the public at all. Competitors who kept their most powerful models behind government-only access walls faced no equivalent action. ## Partial Restoration (June 27) After two weeks of negotiations involving Anthropic co-founder Tom Brown and senior administration officials, Commerce Secretary [[Howard Lutnick]] signed a letter on June 27, 2026, permitting [[Claude Mythos 5]] access for approximately 100 trusted U.S. companies and federal agencies. Lutnick wrote that he had "determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model." The restoration covered [[Claude Mythos 5]] only — not [[Claude Fable 5]], which remained suspended. The ironic result: the restricted model came back before the public one. ## Why It Matters The directive established several precedents the KB tracks: **First AI export control recall.** Previous export controls on AI targeted chips (the October 2022 and January 2025 semiconductor restrictions on China) or pre-deployment licensing. This was the first time the government recalled a deployed commercial model from existing customers based on a discovered vulnerability. **Voluntary safety as liability.** Anthropic designed Fable 5's guardrails as a responsible-deployment mechanism — the very thing the [[Dual-Use AI]] debate says companies should do. When one guardrail was found bypassable, the government treated it as grounds for total suspension rather than acknowledging the system's 95%+ effectiveness. The incentive this creates: don't release powerful models publicly, because public availability is where jailbreaks get discovered and reported. **Ad hoc executive action.** No statute specifically authorizes AI model export control recalls. The directive was issued under general national security authorities, reinforcing the pattern — visible across the [[Anthropic Government Ban]] and the broader Trump administration tech posture — of executive discretion substituting for legislative framework. **Continuation of the Anthropic collision.** This was the second major government action against Anthropic in four months, following the February 2026 [[Anthropic Government Ban]] from Pentagon contracts for refusing to remove military-use restrictions. Different legal basis (export control vs. contract compliance), same company, same underlying tension: Anthropic's safety commitments repeatedly putting it at odds with an administration that views AI restrictions as obstacles to national power. ## Development Log | Date | Development | Source | |------|-------------|--------| | June 12, 2026 | Export control directive issued at 5:21 PM ET; Anthropic disables Fable 5 and Mythos 5 for all customers | CNBC, Anthropic statement | | June 13, 2026 | Anthropic publicly disagrees with directive; Bloomberg and Al Jazeera report on scope | Bloomberg, Al Jazeera | | June 27, 2026 | Lutnick signs partial restoration — Mythos 5 for ~100 trusted organizations; Fable 5 remains suspended | Fortune, CNBC, Semafor | ## Next Steps to Track - Whether Fable 5 access is restored and under what conditions - Whether similar directives are issued against other frontier model providers - Whether Congress acts on AI-specific export control legislation - Impact on Anthropic's IPO timeline (confidential S-1 filed June 1, listing expected ~October 2026) ## See Also - [[Claude Fable 5]] — the publicly available model that was jailbroken - [[Claude Mythos 5]] — the restricted model partially restored June 27 - [[Anthropic]] — the company at the center of the directive - [[Anthropic Government Ban]] — the February 2026 Pentagon contract ban (different legal basis, same company) - [[Dual-Use AI]] — the policy framework the directive exposes gaps in - [[Howard Lutnick]] — the Commerce Secretary who signed the partial restoration - [[Project Maven]] — the Pentagon AI program from which Anthropic was removed - [[Splinternet]] — the broader fragmentation of technology access along national lines