The withdrawal of Anthropic’s AI models, Claude Fable 5 and Claude Mythos 5, is reportedly linked to Amazon’s actions. Reports indicate that Amazon is being identified as the entity that instigated the block. According to sources, the global removal of these models was not a precautionary measure by Anthropic, but a direct result of an escalation involving Amazon, the White House, and Anthropic. Axios reports that Amazon alerted the U.S. Government to a method discovered for circumventing certain model protections and obtaining valuable information for potentially sensitive cybersecurity tasks. This alert triggered an urgent response within the U.S. Administration, which ultimately imposed export controls on Anthropic’s most advanced AI models.
This situation is particularly noteworthy because Amazon is a significant player in this narrative. Amazon is one of Anthropic’s major partners and investors, having announced an immediate investment of $5 billion in April, with an option for an additional $20 billion. This is alongside a strategic relationship where Anthropic committed to spending over $100 billion on Amazon Web Services (AWS) infrastructure over the next decade, and securing up to 5 GW of capacity with their Trainium chips. Therefore, one of Anthropic’s main commercial allies may have also been the source of the information that precipitated the block.
Anthropic Ultimately Withdrew Claude Fable 5 and Mythos 5 Due to Inability to Guarantee Block on Foreign Nationals
Anthropic’s official stance is that they received a directive from the U.S. Government to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States. This included Anthropic’s own foreign employees. As the company could not guarantee practical and secure segregation between U.S. and non-U.S. users, they proceeded to deactivate both models for all their clients.
The technical crux of the model’s removal lies in the term “jailbreak.” The U.S. Administration considers that Amazon demonstrated a way to bypass the protection measures of Fable 5 or Mythos 5 and access capabilities related to vulnerability discovery. Anthropic, however, maintains that a universal jailbreak was not demonstrated, but rather a limited technique that would have allowed the identification of a small number of minor and already known vulnerabilities. The company also asserts that other public models can perform similar tasks without needing to circumvent specific protections.
According to Anthropic, Fable 5 and Mythos 5 were models of a higher class than Opus. Fable 5 was the generalist version with safeguards, while Mythos 5 shared the same base model but with lifted restrictions in specific areas for cybersecurity partners within controlled access programs. Anthropic presented it as the AI model with the highest cybersecurity capabilities on the market, designed to assist defenders in finding and fixing flaws in critical software. However, in the wrong hands, it serves the exact opposite purpose: finding vulnerabilities with ease.
The U.S. Government Interpreted These Capabilities as a Potential National Security Risk
According to Reuters, Anthropic had previously warned about Mythos’s hacking capabilities and had limited its deployment before releasing Fable 5 as the public version with these protective measures. In addition, the relationship between Anthropic and the Trump Administration was already strained due to the company’s refusal to allow certain uses of its models in mass surveillance or autonomous weapons. The report also notes that some experts do not consider what Amazon demonstrated to be a true jailbreak in the strict sense.
Business Insider adds further context to the hours preceding the ban. It indicates that senior White House officials had attempted to persuade Anthropic to voluntarily withdraw the model, while Anthropic’s CEO argued that the security measures were still functioning and that the alleged bypass did not equate to fully opening the model. The Administration, on the other hand, interpreted Anthropic’s response as a lack of urgency regarding a risk they considered grave.
The most immediate consequence is that Claude Fable 5 and Mythos 5 have been taken out of service globally, not just for China, Russia, or other countries considered adversaries, but also for U.S. allies. This means European companies, international researchers, and Anthropic’s own foreign employees are affected. In this way, advanced AI is beginning to be treated not just as software, but as a strategic technology comparable to semiconductors, lithography equipment, or defense systems, thus suffering the same blockades that the hardware industry already faces.
