Science & Tech
U.S. Blocks Anthropic's New Claude Models Over Security Concerns
The government ordered Anthropic to shut off access to Fable 5 and Mythos 5 for all foreign nationals, forcing a global service disruption.
The government ordered Anthropic to shut off access to Fable 5 and Mythos 5 for all foreign nationals, forcing a global service disruption.
’s newest Claude models became a national-security flashpoint when U.S. authorities ordered the company to cut off access on export-control grounds. The dispute turns on whether advanced AI systems can be safely distributed beyond tightly controlled users, with saying it had already added restrictions for while arguing the government’s concern rested on a misunderstanding.
Reporting and Anthropic’s own statement clarified that the order applied to both inside and outside the United States, and not just to users abroad. That meant the company could not simply block one geography, it had to cut off access across its entire customer base.
The restriction applied to foreign nationals inside and outside the United States, not just users abroad. That meant a simple geographic block would not solve the compliance problem, because the company had to exclude a class of users rather than a location.
It affected foreign nationals across Anthropic’s entire customer base. Because the restriction was not limited to overseas users, Anthropic said it had to cut off access broadly instead of screening by region.
Anthropic said the government’s action was tied to and to a suspected way of bypassing or . The company said the government had pointed to a narrow jailbreak claim, while Anthropic argued the technique involved only previously known .
Anthropic said the order was tied to national-security authorities and a suspected jailbreak or bypass of Fable 5. The concern appears to have been that the models could be accessed in ways that circumvented the restrictions the company had added.
Anthropic said the government was relying on a narrow jailbreak claim, while the company believed the technique involved only previously known minor vulnerabilities. That difference matters because it affects how serious the security risk appears and how broad any response should be.
Because it said ordinary service could not reliably enforce the , Anthropic disabled Fable 5 and Mythos 5 for all customers worldwide. The company said access to its other Claude models was not affected, making the shutdown a targeted but .
Anthropic said its normal service controls could not reliably enforce the foreign-national restriction, so it had to disable both models for everyone. In practice, that made the restriction global rather than geographic.
Anthropic said its other Claude models were not affected. The impact was therefore limited to Fable 5 and Mythos 5, but the disruption still reached all customers who depended on those specific models.
Anthropic said it received the directive on Friday afternoon and responded that it had to disable both models immediately to remain . The company said it believed the government action reflected a misunderstanding and that it was working to restore access as soon as possible.
Anthropic said it needed to disable Fable 5 and Mythos 5 immediately to stay compliant with the directive. The company also indicated it viewed the order as based on a misunderstanding, which suggests it planned to seek a reversal or clarification while the models were offline.
Anthropic said it was working to restore access as soon as possible. That implies the company was trying to resolve the government’s concern rather than accept the shutdown as permanent.
On Friday, the U.S. government, citing , ordered Anthropic to suspend access to Fable 5 and Mythos 5 for any , whether inside or outside the United States. The directive also covered Anthropic employees and was framed as an on advanced AI models.
The directive applied to any foreign national, whether inside or outside the United States, and also to foreign national Anthropic employees. That made it much broader than a simple overseas sales restriction.
Anthropic would have to cut off access to Fable 5 and Mythos 5 for foreign nationals covered by the order. In practice, that limits who can use or receive the models and shows the government viewed advanced AI access as a national-security issue.
Anthropic indicated that Mythos 5 would first go only to previously approved organizations through , a government-linked preview program. The company later said it intended to expand access through a for cybersecurity organizations and biology researchers, underscoring that the strongest model was not meant for open distribution.
Anthropic said the first users would be organizations already approved through Project Glasswing, a government-linked preview program. That means access was meant to stay limited to vetted entities rather than open public distribution.
Anthropic said it planned to expand access to cybersecurity organizations and biology researchers through a trusted access program. That would have broadened use, but still kept the model behind a gate instead of making it generally available.
Anthropic said Fable 5 had strong performance in , knowledge work, vision, analytics, scientific research, and -related tasks, but added safety controls in high-risk areas such as , , , and distillation. The company also said Mythos 5 would initially be limited to organizations already approved for its Mythos preview work and that it planned a broader trusted access program later.
Anthropic said it added safety controls for cybersecurity, biology, chemistry, and distillation. The aim was to reduce the chance that the model could be used to help with dangerous technical workflows, even while keeping it useful for normal work.
Anthropic said Mythos 5 would initially be limited to organizations already approved for its previous preview work, with broader access only later through a trusted program. That approach suggests the company believed the strongest version needed tighter vetting before wider release.
Anthropic released Claude Fable 5 and Claude Mythos 5, presenting Fable 5 as its flagship public model and Mythos 5 as a higher-access configuration for specialized users. The launch immediately put the models in a sensitive policy category because Anthropic framed them as unusually capable systems.
Anthropic presented both models as unusually capable, which put them in a sensitive category from the start. When a company frames a model as especially powerful, regulators are more likely to scrutinize who can access it and whether export rules should apply.
Fable 5 was Anthropic’s public flagship model, while Mythos 5 was positioned as a higher-access configuration for specialized users. That split signaled that Anthropic was already treating Mythos 5 as the more controlled and sensitive system.