The ATTCK-R method outperforms other jailbreak attacks on 9 LLMs
A prompt-based safety layer cut the success rate of a strong baseline attack from 0.896 to 0.608.
Claimed, not confirmed
Researchers show ATTCK-R, a framework for multi-turn red-team tests of protected black-box LLMs. They did tests on 9 LLMs, each with one prompt-based safety layer. ATTCK-R outperforms recent multi-turn jailbreaking frameworks. The safety layer cuts the average success rate of a strong baseline from 0.896 to 0.608. The researchers say that tests with no protection show a risk that is too high.
Sources
Posted
