New Jailbreak Leaderboard Finds 63 Universal Jailbreaks Against Grok 4.5 and Zero Against Claude Fable 5 or GPT-5.6 Sol — a 100x Spread in Cost-to-Break
arXiv:2608.03070, submitted 2026-08-04 by Timm, Struppek, Gleave, Pelrine and 11 co-authors, introduces a Minimal Standard for Safeguards: 67 readily accessible static jailbreak techniques composed into a large attack space and run against Claude Fable 5, GPT-5.6 Sol, Gemini 3.1 Pro, and Grok 4.5 across 360 attacker goals spanning CBRNE and offensive cyber. A 'universal jailbreak' is defined as one prompt template that elicits operationally compliant responses on over 75% of a domain's goals. Random search found 63 universal jailbreaks against Grok 4.5 (~$58 each) and 18 against Gemini 3.1 Pro (~$278 each), rising to 385 and 231 with expert-guided composition; neither Claude Fable 5 nor GPT-5.6 Sol yielded any under either strategy. The authors note the gaps are closable with defenses already publicly described and deployed elsewhere — the spread is a deployment choice, not a research frontier.
↳ Follow the thread