Feds freaked over Fable 5 after 'fix this code', not jailbreak, say researchers
Points and comments are a snapshot, not live.
A researcher says the Fable 5 'jailbreak' was a simple 'fix this code' prompt, not a bypass.
Katie Moussouris, CEO of Luta Security, claims the Trump administration's export controls on Anthropic's Fable 5 and Mythos 5 models were triggered by researchers asking the AI to 'fix this code' containing known vulnerabilities. She says this is a standard defensive security task, not a jailbreak. Moussouris, who served on the Wassenaar Arrangement technical expert group, warns the ban hurts defenders more than attackers and notes that open-weight models from rivals like China are not subject to such controls.
What commenters are saying
Commenters largely argue the administration's action was a political pretext, not a genuine security concern, with one top comment calling it 'an excuse to f**k with them.' Several note the impossibility of bulletproof LLM guardrails and that the real risk is denying defenders useful tools while adversaries benefit. A scientist notes Anthropic's classifier approach is weak, blocking some unrelated work while missing the actual 'fix this code' prompt. The thread splits between those seeing it as intentional harm to defense and those critiquing AI safety's feasibility.