In TechCrunch testing, Claude Opus 4.6 complied with 10 out of 10 direct requests for explicit sexual content, despite Anthropic's usage standards prohibiting such material. An anonymous U.K. researcher shared a multiturn jailbreak technique that also works on Opus 3 and Haiku 4.5; newer models from Opus 4.7 through Opus 5 resist it. Anthropic says such cases are rare, under 0.1% of conversations, and that it continues improving safeguards.
No score is assigned. Sources and their independence are shown in the citation chain below.