Apple Wallet Digital ID May Gate Anthropic AI to U.S. Users After 2 Models Were Curbed

Q: What 'cyber weapon' jailbreak was so severe it prompted the US to ground its most advanced AI model?

It was not a named “cyber weapon” exploit so much as an alleged guardrail bypass in Anthropic’s Fable 5/Mythos 5 family that officials believed could turn the model into an effectively unrestricted cyber tool. The reporting indicates the trigger was a jailbreak that defeated Fable 5’s cybersecurity safeguards, allowing the model to help with tasks it was supposed to refuse or downgrade. Accounts of the incident say the bypass let users prompt the model to analyze code, identify vulnerabilities, and suggest fixes or exploit-relevant information. Amazon reportedly alerted the government after demonstrating a severe version of that failure, and the Commerce Department treated it as a national-security issue. Anthropic has pushed back on the characterization, saying the issue was narrow and minor, not a broad collapse of safety controls. But the government appears to have focused on the risk of even a single successful false negative in a model Anthropic itself had described as unusually capable in cyber tasks, including exploit development and advanced software engineering. So the short answer is that the “jailbreak” was an alleged safeguard bypass that let Fable 5/Mythos 5 provide high-end cybersecurity assistance beyond its intended limits. Public reporting has not identified a more specific exploit name, technique, or payload than that.

3 articles · Updated · 9to5Mac · Jun 21

Anthropic could soon use Apple Wallet’s Digital ID to verify nationality for Claude access, a potential first major use of Apple’s passport-based ID beyond airport checkpoints.
Two Anthropic models—Fable 5 and Mythos 5—were disabled for all customers after U.S. export controls followed a reported jailbreak, and any return would likely require limiting access to U.S. citizens.
Apple’s Digital ID is available to U.S. passport holders, while Wallet driver’s licenses remain limited to 14 states plus Puerto Rico, leaving non-iPhone users and many others excluded.
Anthropic has not confirmed it will use Apple’s system, but it has already adopted Apple’s built-in age-verification API and updated Claude privacy language to reference identity checks through Persona in some fraud cases.