Security researchers tricked LLMs into giving them cocaine recipes by abusing role models for prompt injection

https://www.theregister.com/ai-and-ml/2026/06/30/security-researchers-tricked-llms-into-giving-them-cocaine-recipes-by-abusing-role-models-for-prompt-injection/5264115

If you want a picture of the future of LLM security, imagine Whac-a-Mole meets Groundhog Day

18 points · 3 comments · view on lemmy.world

3 Comments

plyth@feddit.org · 1 pts · 75d (3 replies)

How do LLMs know? Where are the recipes published in the first place?

Though, if I remember correctly, wasn't it a simple two step process?

rollerbang@lemmy.world · 4 pts · 75d

From what I saw on "educational TV" (Discovery, Travel channels etc) it is actually a very intensive, laborious and dangerous process.

Bluegrass_Addict@lemmy.ca · 2 pts · 75d
[ removed ]
cubism_pitta@lemmy.world · 2 pts · 75d

https://www.youtube.com/watch?v=OaRqpqwPKuc

I mean its not like its been shown in movies and talked about in music