Anthropic shuts down Fable 5, Mythos 5 as US govt shuts down foreign national access

The US government cited national security concerns and suspended access to any foreign national using Claude Fable 5 and Mythos 5 inside or outside the US. Anthropic then moved to disable all customer access to Fable 5 and Mythos 5 to ensure compliance. Welcome to the new unknowns of frontier model releases.

Anthropic's move was designed to ensure compliance and the rest of its models are available. Perhaps talking so much about the dangers of Anthropic's models backfired.

• Anthropic’s Claude Fable 5, Mythos 5: What you need to know
• Anthropic backtracks on covert model switching for Claude Fable 5
• Anthropic on regulation and policy: We need to keep up
• Fable 5 and the Shift From Model Capability to Capability Governance
• My internal monologue reading Anthropic's big idea posts
• Claude Mythos and the New Cybersecurity Operating Model

In a blog post, Anthropic said:

"We received the directive from the government today at 5:21pm (ET). The letter did not provide specific details of its national security concern. Our understanding is that the government believes it has become aware of a method of bypassing, or “jailbreaking” Fable 5. We reviewed a demonstration of this specific technique being used to identify a small number of previously known, minor vulnerabilities. These vulnerabilities all appear relatively simple, and we have found that other publicly available models are able to discover them as well without requiring a bypass."

Anthropic detailed its safeguards, which were annoying to researchers. Anthropic then moved to more transparent safeguards so customers knew what was going on.

The company's beef with US government decision is that it worked with officials leading up to the Fable 5 launch, said its defense strategy is strong and that concerns about a jailbreak that's universal is overblown. "Perfect jailbreak resistance does not appear to be possible today," the company said.

Anthropic added:

"We have not even received a disclosure of a concerning non-universal potential jailbreak that led to a harmful result. The potential jailbreaks that have been disclosed to us are either entirely benign responses or are minor findings that provide no Mythos-specific uplift."

A few thoughts:

• First, a US model recall is unprecedented and there are a lot of moving parts to this one.
• Model recalls may become the norm.
• This could all be resolved by Monday.
• OpenAI should be taking notes and it's unclear how its next model launch will fare.
• Anthropic's security warnings appear to have backfired.
• Sovereign AI efforts are going to accelerate as no country can have an LLM just pulled.
• It's hard to see Anthropic or OpenAI going public with this level of launch uncertainty. At the very least the S-1 risk factor should be interesting.
• Anthropic appears to be getting its wish that AI should be slowed down to ponder bigger implications.
• Can this move pop the AI bubble?