Anthropic Reinstates Fable 5 and Mythos 5 Following Lifted U.S. Export Restrictions

TLDR

• Export restrictions on Anthropic’s Fable 5 and Mythos 5 were removed by U.S. authorities on June 30
• The ban began June 12 following discovery of a jailbreak exploit revealing security weaknesses
• Unable to verify user nationalities instantly, Anthropic blocked all users from accessing the models
• The company committed to intelligence sharing and providing early government access to new models
• Rising competition from Asian AI systems like Fugu and Tulongfeng influenced U.S. policy shift

Anthropic has begun reinstating worldwide access to its most sophisticated AI systems following the removal of U.S. government export restrictions that had rendered them unavailable throughout much of June.

Global availability of Fable 5 resumed July 1 across all Anthropic services. Mythos 5, which operates on identical architecture but features relaxed safety guardrails, is being made available to a select cohort of U.S.-based organizations that secured governmental clearance on June 26.

The Genesis of the Blockade

The regulatory action commenced June 12 when federal authorities designated both systems as export-controlled technology. Under this classification, distribution to individuals of foreign nationality required explicit authorization.

A cybersecurity discovery precipitated the decision. Researchers from Amazon uncovered a jailbreak technique that circumvented safety mechanisms, enabling Fable 5 to detect software security flaws and generate exploit code demonstrating potential attack vectors.

Faced with immediate enforcement and lacking infrastructure to authenticate users’ citizenship status in real time, Anthropic withdrew access universally to avoid regulatory violations.

Security professionals questioned the restrictions’ validity from inception. They observed that Anthropic had already committed to implementing comparable security protocols voluntarily well before export regulations materialized. Some industry observers interpreted the prohibition as a political tool rather than a genuine security intervention, potentially connected to Anthropic leadership’s public disagreements with Trump administration AI governance strategies.

Anthropic’s Commitments for Restoration

Commerce Secretary Howard Lutnick disclosed that Anthropic pledged to actively identify and mitigate security vulnerabilities, establish joint protocols with government entities for future deployments, and report any detected malicious exploitation attempts to federal authorities.

The company is expanding its partnership framework with government departments. This encompasses providing designated agencies with pre-release access to cutting-edge models ahead of public launches and exchanging threat intelligence data.

Anthropic is developing a standardized assessment system alongside Amazon, Microsoft, and Google to evaluate the severity level of discovered jailbreaks. These collaborative measures align with a June 2 executive directive addressing AI security infrastructure.

International Market Dynamics Influenced Decision

AI developers across Asia have started deploying systems with capabilities rivaling Mythos-level performance, notably Fugu and Tulongfeng. This competitive landscape created strategic complications for U.S. policymakers, undermining the rationale for restricting American AI technologies from international distribution.

Lutnick also authorized Mythos access for specific White House-vetted clients last week. OpenAI’s most recent systems followed a comparable distribution strategy, reaching only pre-approved entities rather than broad public availability.

While Anthropic remains privately owned, market participants have speculated on its valuation through pre-IPO derivative instruments on Hyperliquid’s blockchain-based exchange. That contract declined approximately 3.7% when access restrictions were implemented in June.

The Trump administration’s evolving AI regulatory framework has created ambiguity throughout the technology sector regarding compliance expectations for upcoming model deployments.