Claude Fable 5 remained offline for a fifteenth day as the United States allowed Anthropic to restore limited access to Claude Mythos 5 for vetted organisations involved in critical infrastructure and cyber-defence work.The partial reversal, cleared on June 26, eased pressure on hospitals, utilities, financial institutions and government-linked security teams that had been cut off from Anthropic’s most powerful cyber-focused model after an export-control directive forced the company to disable both Fable 5 and Mythos 5 worldwide. The consumer-facing Fable 5 model, however, has not been brought back online, leaving developers, enterprises and paying users without a firm timetable for service restoration.
The directive, issued on June 12, barred access to the two models by foreign nationals, including people working inside the United States and employees of Anthropic itself. Because Anthropic said it could not reliably verify every user’s citizenship in real time across its products and application programming interfaces, the company suspended access broadly to avoid violating the order. Other Claude models were left unaffected.
Commerce Secretary Howard Lutnick’s June 26 authorisation created a narrow path for Mythos 5 to return through approved providers and institutions. The clearance applies to more than 100 trusted organisations, with priority given to entities responsible for energy systems, telecommunications networks, transport infrastructure, financial stability, health services and government cyber operations. The arrangement also reduces the need for case-by-case export licences for approved users, including non-citizen staff working within cleared organisations.
The continued absence of Fable 5 shows the limits of the policy shift. Fable 5 had been marketed as Anthropic’s most capable widely released model, aimed at advanced coding, research, design, autonomous workflow and general productivity tasks. Mythos 5, by contrast, sits inside a more restricted programme focused on cybersecurity and high-assurance use cases. That distinction appears central to the government’s decision to restore one model while keeping the other under tighter control.
The restrictions followed escalating concern in Washington that frontier AI systems could be adapted for offensive cyber operations, automated vulnerability discovery or misuse by hostile states. Officials have been weighing how to apply export-control principles, traditionally used for chips, software and dual-use equipment, to AI models that can be accessed remotely and reproduced through cloud services.
Anthropic introduced Fable 5 and Mythos 5 on June 9, presenting them as major advances in reasoning, coding and agentic task execution. Fable 5 was positioned for broad commercial deployment, while Mythos 5 was placed behind an invitation-only framework for sensitive cyber-defence work. Within days, the launch had turned into a test case for how governments may intervene when model capability is judged to cross national-security thresholds.
The shutdown drew sharp criticism from cyber professionals who argued that denying access to defensive AI tools could weaken rather than strengthen security. Teams responsible for threat detection, incident response and vulnerability management had already begun building workflows around the models. Some warned that attackers would not wait for regulatory clarity, while defenders would lose access to capabilities that helped them analyse malware, triage alerts and test systems faster.
Supporters of the directive counter that the same capabilities can be used to accelerate exploitation, automate reconnaissance and lower the skill threshold for cyber operations. The central question is whether risk should be managed by restricting access at the model level, controlling deployment environments, auditing users, or imposing obligations on providers before release.
Academic work on frontier-model safety has added weight to both sides of the debate. Red-team studies of advanced systems have found that even hardened models can be pushed into producing harmful outputs under sustained automated attack. Other research has stressed that risks may lie not only in model behaviour but also in the surrounding containment infrastructure, including sandboxes, tool permissions and cloud execution environments.
The Fable 5 suspension has also intensified concern among international customers. A directive aimed at foreign-national access effectively affects multinational companies, universities and research labs whose teams operate across borders. The issue is particularly complex for AI firms, where top researchers, engineers and security specialists often work in mixed-nationality teams.
Anthropic has said it is working with the government to expand access while preserving compliance with national-security requirements. The company is expected to prioritise restoration for enterprise and public-sector users before any broader return of Fable 5. Developers affected by the blackout have shifted workloads to older Claude models or rival systems, though many report performance gaps on complex coding and autonomous tasks.
Topics
Technology