logoalt Hacker News

Yizahi • today at 1:31 PM • 1 reply • view on HN

Root cause - guardrails "unexpectedly" were neither guards nor rails. Who knew that asking a stochastic program "pretty please" is not a strict instruction or a barrier whatsoever.


Replies

fabioborellini • today at 3:42 PM

Bet there was an executive telling we cannot afford old-time security in the era of AI. If they had guardrails outside of the prompt they were getting left behind.