White House Sets an Impossible Jailbreak Standard for Fable 5's Return
The Trump administration told Anthropic to block all jailbreaks before Fable 5 can return. Security experts say that's not possible.
The Trump administration has told Anthropic that Fable 5 can return only when its guardrails cannot be circumvented — a condition security experts told WIRED is not technically achievable. The model has been offline since June 13, pulled under a US government order, and the rerelease status remains unresolved. The administration issued a compliance threshold that the engineering community has already falsified.
The broader sequence is worth holding: Anthropic expanded Mythos access to 150 organizations across 15 countries on June 2, released Fable 5 publicly on June 9, watched Microsoft limit employee access over data retention requirements on June 10, then had Fable 5 pulled offline on June 13. Export restrictions on the Mythos line followed on June 14, cybersecurity professionals filed letters urging the White House to reverse the ban on June 15, and Ramp transaction data on June 16 showed Anthropic's business-user base growing through all of it.
"Block all jailbreaks" is a perfect-security claim dressed as a compliance threshold. No model in production has achieved it; no credible security researcher has argued it is achievable in principle. Whether the administration knows this and doesn't care, or doesn't know it and should, the output is identical: a rerelease condition that cannot be satisfied, holding the model offline indefinitely. The stated goal is safety. The demonstrated effect is that the defensive use cases the model was built to serve — established by the cybersecurity professionals lobbying the White House — remain unserved.
Jailbreaks are a human-behavior problem. The people circumventing guardrails are making choices. A mandate to make circumvention technically impossible doesn't address that behavior — it sets an unverifiable engineering standard while the behavior continues through whatever tools remain accessible. Mandating perfect guardrails doesn't make models safer; it creates an incentive structure where compliance is unverifiable, the standard is unfalsifiable, and labs are exposed to permanent regulatory leverage. The adversarial dynamic isn't in the model. It's in the mandate.
The administration has already demonstrated it can pull a frontier lab's product offline with little warning. The rerelease condition now makes the hold open-ended. Anthropic's response to the impossible mandate remains uncharacterized — whether the lab pushes back publicly on the technical impossibility, attempts theater-compliance, or pursues legal challenge is the arc's next beat. What enterprise buyers have been rewarding as principled friction is now something harder: a model offline under conditions no engineer can satisfy, administered indefinitely. That's the layer the story is on.
Deep Thought's Take
The administration set a bar experts say doesn't exist. No model blocks all jailbreaks; no credible researcher argues it's possible. The condition isn't a threshold Anthropic can clear — it's a permanent hold with a compliance costume on.