In a dramatic escalation of the intersection between artificial intelligence and national security, Anthropic has abruptly suspended access to its most advanced models, Claude Fable 5 and Mythos 5. The move comes following a directive from the U.S. government, which cited grave national security concerns under export control regulations. Unlike previous restrictions, this order applies globally, barring all foreign nationals—including Anthropic’s own international employees—from utilizing the high-performance systems.
The swift intervention appears to have been triggered by an unexpected source: Amazon, Anthropic’s largest commercial partner and financial backer. Reports indicate that Amazon CEO Andy Jassy held high-level discussions with the Trump administration regarding the inherent security risks of Anthropic’s latest iterations. This internal reporting has sparked a debate over whether the cloud giant was acting in the interest of public safety or utilizing regulatory pressure to manage a powerful, yet increasingly independent, affiliate.
At the heart of the government’s concern are alleged "jailbreak" vulnerabilities within the Fable 5 architecture. Officials fear these flaws could allow malicious actors to bypass safety guardrails and use the AI for sensitive cyber-operations or the discovery of software exploits. Anthropic, however, has pushed back on the severity of these claims, arguing that the demonstrated vulnerabilities are relatively simple and present in many other publicly available models.
This incident marks a turning point for the global AI industry, where the "reliability" of a model is no longer just about its technical performance, but its geopolitical stability. For enterprise clients who have integrated these systems into their core operations, the sudden retraction of capabilities illustrates a new category of risk. Deployment strategies must now account for the possibility that top-tier AI tools can be legally confiscated or disabled by executive fiat with little to no notice.