Researchers gaslit Claude into giving instructions to build explosives

Corroborated by 2 sources from 2 publishers

globalscience4h ago

TL;DR

Reports differ across sources; according to theverge.com, anthropic has spent years building itself up as the safe AI company.

Sources