White House Demands Anthropic Fix Jailbreak Vulnerabilities Before Fable 5 Re-release; Experts Say It's Nearly Impossible
Claire Weston
The White House told Anthropic to patch Claude Fable 5's jailbreak vulnerabilities before the model can go back online, but independent security experts say AI guardrails are inherently temporary — a permanent fix may not exist.
Why was Fable 5 taken offline?
The U.S. government pulled Claude Fable 5 last week under export-control authority, after finding that users could bypass safety guardrails via jailbreak prompts.
Those guardrails were designed to block access to cybersecurity, chemical, and biological capabilities inside the Mythos model.
This means → the concern is not the model's raw performance — it is that the safety lock can be picked.
What are the two sides saying?
Anthropic has argued for days that the government's concerns are overblown and that jailbreak impact is limited. The company repeated that position Monday in a technical meeting with the Commerce Department and National Cyber Director Sean Cairncross's office.
The government has shifted its stance: it is no longer debating the severity of the flaw — it is telling Anthropic to fix it.
The NSA has determined that methods exist to disable Fable 5's guardrails.
In plain terms = the government's message is: "We're past the question of whether the hole is real. Patch it."
Why isn't the government patching it itself?
Three people familiar with the discussions say the Commerce Department's AI Standards and Innovation Center and the NSA lack the staff and bandwidth to track every possible jailbreak across every model on the market.
The government therefore expects Anthropic to proactively and continuously test Fable 5 — and all frontier AI models — for jailbreak paths, and self-report findings.
This reflects a larger reality: U.S. frontier-AI safety oversight depends heavily on companies policing themselves.
Can jailbreaking actually be fixed for good?
Independent cybersecurity experts increasingly lean toward one conclusion: AI guardrails are a temporary fix at best.
The reason is straightforward — skilled users and future, more capable AI models can still find ways around restrictions.
This means → the White House is demanding Anthropic "fix it before relaunch," but experts believe the finish line the government is pointing to may not exist.
Content is for reference only, not financial advice.