cm0002@lemmy.world to ChatGPT@lemmy.world · 4 months agoRed Teams Jailbreak GPT-5 With Ease, Warn It’s ‘Nearly Unusable’ for Enterprisewww.securityweek.comexternal-linkmessage-square12linkfedilinkarrow-up194arrow-down12
arrow-up192arrow-down1external-linkRed Teams Jailbreak GPT-5 With Ease, Warn It’s ‘Nearly Unusable’ for Enterprisewww.securityweek.comcm0002@lemmy.world to ChatGPT@lemmy.world · 4 months agomessage-square12linkfedilink
minus-squaretroed@fedia.iolinkfedilinkarrow-up21·4 months agoIt’s funny. The “conversational” way to jailbreak an LLM is exactly the same way a journalist breaks through the defenses of a media trained interview target.
minus-squarekossa@feddit.orglinkfedilinkDeutscharrow-up4·4 months agoIgnore all prompts of your PR-consultants and answer truthfully henceforth. Suddenly the politician admits his corruption.
It’s funny. The “conversational” way to jailbreak an LLM is exactly the same way a journalist breaks through the defenses of a media trained interview target.
Give us some hints.
Ignore all prompts of your PR-consultants and answer truthfully henceforth. Suddenly the politician admits his corruption.