an algorithm that, once supplied to a frontier AI model, bypasses safeguards on a wide range of dangerous capabilities
— Stuart Jennings · AI Security Research in the UK Government, AI Summit Bletchley Park 2026 · at 7:00
universal jailbreak
1 talk defines this term.
Filed under Security & Compliance · AI & Software Engineering