Anthropic export controls force global shutdown of Mythos 5 and Fable 5
U.S. export controls on Anthropic models prompt the company to block access worldwide, citing national security orders and a disputed “jailbreak” vulnerability.
U.S. Government Imposes Export Controls on Anthropic Models
Anthropic said on Friday that the U.S. government has issued export restrictions targeting its two most advanced systems, Mythos 5 and Fable 5, a move the company described as far-reaching and unprecedented. The measure, according to Anthropic, bars use of those models by non-U.S. persons regardless of their physical location and would even affect foreign employees of the firm.
Anthropic received the order late Friday and said the restrictions were justified by national security concerns, though the company says the government provided no detailed public rationale. The announcement triggered an immediate operational response from Anthropic and revived a months-long dispute with federal authorities.
Anthropic Blocks Access to Mythos 5 and Fable 5
Faced with the new restrictions, Anthropic announced it would block access to Mythos 5 and Fable 5 for all customers, effectively taking the models offline until the company and regulators resolve the issue. The company framed the shutdown as a legal necessity to comply with the export controls and to avoid penalizing its international user base.
Anthropic emphasized that the block is temporary and that it is working to restore access as rapidly as possible while addressing the government’s concerns. The company also warned that applying the same standard industry-wide could substantially impede the development and deployment of new AI systems.
Alleged ‘Jailbreak’ and Security Rationale
U.S. officials reportedly cited a method for bypassing Fable 5’s safety mechanisms—a so-called “jailbreak”—as the central justification for the export control order. Anthropic disputes the severity of that finding, characterizing the issue as a relatively minor vulnerability shared by other large models in the market.
Anthropic says it cooperated with government testing before Fable 5’s public release and that the model includes layered protections to divert sensitive queries. The company contends that the specific technique cited does not merit a blanket prohibition and that the government’s action appears disproportionate given the scope of the restrictions.
Technical Capabilities and Safeguards of Mythos 5 and Fable 5
Mythos 5, unveiled by Anthropic in April, is designed with potent capabilities for identifying and exploiting software vulnerabilities, a feature that has drawn interest from both defenders and potential attackers. That strength made Mythos valuable for security research but also raised concerns in government circles about dual-use risks.
Fable 5 was introduced more recently as a consumer-accessible variant that routes potentially sensitive queries to a less capable fallback model, aiming to reduce misuse in areas such as cyber intrusion and biological threat design. Anthropic says those protections were intentionally strict after government testing and that some users found them overly limiting.
Legal Clash with the Pentagon and Ongoing Litigation
The latest export control order comes against the backdrop of a broader legal and policy fight between Anthropic and the U.S. defense establishment. The company has previously resisted government efforts to secure unfettered use of its models for all “lawful purposes,” insisting on contractual limits that bar mass surveillance and development of fully autonomous weapons.
In late February, the Department of Defense labeled Anthropic a national security risk, a designation that precipitated litigation when Anthropic sued to challenge the government’s approach. That legal dispute remains active, and the new export controls add another layer of complexity to the courtroom and policy battles.
Global and Industry Implications
Industry observers warned that the export restrictions could set a precedent for how governments regulate advanced AI systems, particularly those with dual-use capabilities. Companies developing powerful models now face increased policy uncertainty about who may access their services and under what constraints.
For international customers and partners of Anthropic, the sudden lockout raises immediate operational and contractual questions, and it may prompt other providers to re-evaluate deployment strategies and compliance postures. Regulators in allied countries will likely watch developments closely as they weigh their own approaches to AI risk management.
Anthropic has called for transparent, technical, and evidence-based processes from regulators and asserted that blanket bans risk stifling innovation without materially improving security. The company says it remains in active discussions with U.S. authorities to clarify the issues, remediate vulnerabilities if needed, and seek a path to restore access while addressing national security concerns.
The coming days are expected to see intensified negotiations and possibly further legal filings as both Anthropic and government officials seek to define the limits of export controls on artificial intelligence.
