Days after unveiling its highly anticipated AI model with Mythos-level capabilities, Fable 5, it appears the model may soon become inaccessible to many users worldwide. On Friday, Anthropic
announced that it was disabling access to its latest models, Fable 5 and Mythos 5, following a US government export control directive requiring the company to suspend access for all foreign nationals over national security concerns.
In its statement, the company said it was ordered to restrict access without being provided an evidence-backed rationale for the US government’s concerns. According to Anthropic, officials believe that there may be a way to bypass or jailbreak the safety mechanism deployed into Fable 5, which potentially allows it to be used to identify software vulnerabilities.
However, the Dario Amodei-led company has disputed the severity of the issue. The company said that it has only been offered what it describes as verbal evidence of a potential narrow, non-universal jailbreak adding that it does not believe such findings justifies recalling a commercial AI model that is already accessible to millions of users.
“The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Anthropic models will not be affected,” Anthropic said in its statement.
When it comes to the implications, the impact of the directive could likely extend beyond the US borders. According to former White House official Dean Ball, the directive implies all non-Americans, including those residing within the US, could be prohibited from accessing Anthropic’s newest models. This also suggests that users may even be required to prove their citizenship.
However, Anthropic maintains that it believes there has been a misunderstanding and that it is working with the authorities to restore access as quickly as possible. Meanwhile, according to a Reuters report, Amazon Web Services confirmed that Anthropic requested it revoke access to Fable 5 and Mythos 5 for users across all regions to comply with the government directive.
The latest directive from the US government marks a shift in its policy, as until now export controls were largely focused on restricting the sale of advanced chips and AI hardware instead of limiting direct access to AI models themselves. According to Anthropic, if this is broadly adopted then such precedence could effectively halt the deployment of new frontier AI systems across the industry.
The order emerges amid ongoing strife between the US government and Anthropic. Earlier this year, the company which is aiming to go public, reportedly refused to allow its AI technology to be used for domestic surveillance and fully autonomous weapons systems. Following the back and forth, Anthropic was placed on a government supply chain blacklist which was scheduled to take effect later this year.
Ironically, a few days ago, Anthropic had publicly called for stronger government oversight of advanced AI systems including mechanisms to block models that pose catastrophic risks. However, the AI startup seems to be arguing that Friday’s directive does not reflect fair or evidence-based regulation.
The latest disagreement follows the recent launch of Claude Fable 5. The model was launched two months after Anthropic claimed it had built an AI model too powerful for wider rollout. Claude Fable 5 which comes with Mythos-level capabilities was introduced along with another AI model called Claude Mythos 5 which the company claims to have broader capabilities than the Mythos Preview Model that was released in April to a closed group.
According to Anthropic, these AI models come with safeguards that prohibit use in sensitive areas such as cybersecurity, even though some users have already criticised these restrictions as overly broad. Meanwhile, security experts too have warned that highly capable AI systems could significantly speed up sophisticated cyberattacks if misused, especially in sectors such as banking, and industries that rely on complex legacy infrastructure.
Upon launch, Anthropic said it was releasing Claude Mythos 5 to a limited group, primarily participants in Project Glasswing, a joint industry initiative to find and fix critical software vulnerabilities using AI. In addition, select biology researchers will be given access to the model, and the company said it will collaborate with the US government on its rollout.
Claude Mythos 5 is the same underlying model as Fable 5, but with some safeguards lifted and initially available only to a limited set of trusted users. By contrast, Claude Fable 5 is being released publicly with guardrails that block it from answering queries related to cybersecurity, biology and chemistry.
When it comes to use cases, Claude Fable 5 is designed for coding, advanced mathematics and visual reasoning. On the performance front, it topped Vals AI benchmark evaluations, outperforming other publicly available models and improving its aggregate benchmark performance by around 5 per cent over Claude Opus 4.8. Anthropic says Fable 5 can work autonomously for longer and is more token-efficient, earning top scores on Cognition’s FrontierCode coding benchmark. However, it is twice as expensive as Opus 4.8 and trails some rivals in healthcare and tax-related evaluations. Fable 5 is available to paid Claude subscribers and developers through Anthropic’s API.