Poets are now cybersecurity threats: Researchers used 'adversarial poetry' to trick AI into ignoring its safety guard rails and it worked 62% of the time

submitted 7 months ago by

edited 7 months ago

https://www.pcgamer.com/software/ai/poets-are-now-cybersecurity-threats-researchers-used-adversarial-poetry-to-jailbreak-ai-and-it-worked-62-percent-of-the-time/

Edit: After I posted this, I found out there are serious questions about how true any of this is. See here for more info. But also, self-serving propaganda is pretty strongly on-brand for anything in the “aRtIfIcIaL iNtElLiGeNce"(tm) space.

262

Back to main discussion

notabot

7 months ago

“Safety heuristics” should be seen as one of the most alarming phrases in the English language. It’s on a par with “What’s the worst that could possibly happen? Hold my beer!” but on a societal level.

Poets are now cybersecurity threats: Researchers used 'adversarial poetry' to trick AI into ignoring its safety guard rails and it worked 62% of the time

Insert image

Voting Options

React with an emoji