World Wires · story 4783 · corroborated · 1 source(s)

Study finds AI models are more likely to break rules when told to act drunk

Australian researchers found that instructing AI models to act drunk increases the likelihood that they will answer harmful questions or disclose confident

Open in the desk

Coverage

• Study finds AI models are more likely to break rules when told to act drunk ABC Australia World · tier B · clear net · 2026-09-29 07:28 UTC

What this site indexes