White House Wants Fable 5 Jailbreak-Proof, Experts Say It Can't Be Done

The standoff over Anthropic's Fable 5 just got a condition attached, and it may be one no company can meet. The White House is reportedly demanding that Anthropic make Fable 5 impossible to jailbreak before the model can be re-released. Security experts say that bar cannot be cleared by any AI company, on any model. So a temporary suspension starts to look like it could become permanent.

Quick recap. On June 12 a US export-control directive pulled Fable 5 and its sibling Mythos 5 for all foreign nationals, after the government said it found a way to jailbreak the model. The trigger, per the Wall Street Journal, was Amazon CEO Andy Jassy flagging the flaw to the White House. The technique is almost mundane: ask the model to read a codebase and find software flaws, which turns its reasoning into a vulnerability-hunting tool. Useful for defenders, dangerous in the wrong hands.

Here is why experts call the new demand impossible. The capability behind the jailbreak is not a bug you patch, it is the model being good at analyzing code. Katie Moussouris, CEO of Luta Security, put it bluntly: the behavior cannot meaningfully be fixed, and any attempt would only weaken the model for defense. Anthropic said the same thing at launch, that perfect jailbreak resistance is not currently possible for any model provider, and built Fable 5 around making attacks narrow, expensive, and closely monitored instead. The same capability sits in GPT-5.5, Gemini, and every open-weight model already downloaded.

There is no Anthropic stock to trade, so the spillover shows up elsewhere. Amazon, whose CEO set the chain in motion and which is also an Anthropic investor, traded near 246 dollars. Nvidia, the chip supplier underneath all of it, sat around 208. The broader AI tape held steady. The real cost here is not a price move, it is the precedent, since a jailbreak-proof-or-banned standard would hang over every frontier model, not just Fable.

That is the part the industry is watching. If the White House holds an unmeetable line, Fable 5 may simply not come back, and the message to every lab is that one disputed vulnerability report can pull a flagship model. David Sacks, the administration's AI adviser, has said he hopes Anthropic fixes it so Fable can return soon, which sits awkwardly next to experts saying there is nothing fixable. The export angle also does little against foreign or open models that have the same capability and sit outside US reach. Critics already argue the ban was never really about the jailbreak.

So Anthropic is being asked to do something the field says is not possible, then judged for not doing it. Build a capable model that cannot be misused, while every rival model can be. The suspension was supposed to be the hard part. The condition for lifting it might be harder. Worth watching whether impossible turns into indefinite.