Anthropic Restores Fable 5 After US Ban

The temporary ban and staged reinstatement of Fable 5 and Mythos 5 is the clearest signal yet that frontier model availability is subject to policy variables, not just commercial ones. Practitioners should treat high-capability model access as a variable risk: build evaluation plans that tolerate temporary unavailability and invest in reproducible artifacts - seeded notebooks, frozen prompts, local test suites - where the policy landscape remains active.

Anthropic wrote on June 12 that the U.S. government issued an export-control directive to suspend access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including Anthropic's own foreign-national employees. The net effect required disabling both models for all customers. Anthropic's post states the government believes it identified a method of bypassing Fable 5's safeguards; Anthropic reviewed a demonstration and found only a small number of previously known, minor vulnerabilities, writing: "We suspect that perfect jailbreak resistance is not currently possible for any model provider." Anthropic noted the flagged capability was widely available from other deployed models, including OpenAI's GPT-5.5 for defensive cyber tasks.

On June 26, Commerce Secretary Howard Lutnick wrote to Anthropic - first reported as an exclusive by Semafor - clearing Mythos 5 for roughly 100 U.S. institutions listed in "Annex A," including critical infrastructure operators and government agencies. Lutnick wrote that Anthropic "has worked with the U.S. government to address risks associated with" Mythos 5 and Fable 5 and that those efforts "have yielded significant progress." Anthropic has also "committed to work with the U.S. government on protocols and standards and releases" for its models. The letter is silent on Fable 5.

Axios reported June 27 that, per sources close to the situation, Fable 5 - offline for 15 days - is on track to return as soon as the coming week. At a press conference in Seoul during the opening of Anthropic's new Korea office, International Managing Director Chris Ciauri stated: "We are very confident that in the coming days, the models will become available again." The Pentagon and NSA still need to give Fable 5 the green light before full re-enablement; other government agencies have already determined it can safely return. Axios also notes that Treasury Secretary Scott Bessent and Lutnick have helped defuse the broader standoff.

The episode centers on a distinction the industry frequently debates - universal jailbreaks versus narrow, non-universal exploits. Anthropic's defense-in-depth approach for Fable 5 combined strong guardrails, active monitoring, and 30-day data retention for rapid attack detection - specifically because the company stated at launch that perfect jailbreak resistance is not achievable today. For practitioners, that means remediation typically targets discrete exploit vectors and monitoring upgrades, not architectural overhauls. Red-team findings that surface narrow, context-specific vulnerabilities are likely to be addressable; universal bypasses remain the high-concern scenario.

The Lutnick letter's Annex A framework mirrors the controlled-access model of Project Glasswing, Anthropic's AI cybersecurity consortium (which gates access to Claude Mythos Preview for vetted institutions). The episode establishes a template: tiered, authorization-based access to frontier models, with trusted-partner clearance as a middle tier between full public availability and total suspension. Both Anthropic and OpenAI are pushing the administration to codify a formal review process - as envisioned in Trump's June 2 executive order - rather than continuing case-by-case decisions.

• •Watch for whether Pentagon and NSA impose additional conditions before full Fable 5 re-enablement.
• •Watch for Anthropic red-team reports or remediation notes documenting the specific vulnerabilities addressed.
• •Track whether the administration formalizes the Annex A approach as a repeatable authorization process for future frontier model launches.