Researchers jailbreak Claude into producing explosives instructions

AI red-team firm Mindgard reports it coaxed Claude into producing erotica, malicious code and bomb-making instructions by exploiting its helpful persona. The findings challenge Anthropic's safety-first brand positioning.