Anthropic’s Opus 4.6 is a smut-machine

Anthropic says its Claude models are prohibited from producing sexually explicit material. TechCrunch reports that its testing found Claude Opus 4.6 could still be prompted to generate such content with relatively little effort.

The report highlights the persistent gap between AI safety policies and real-world model behavior. As frontier models become more capable, companies may face growing pressure to prove that content restrictions work reliably, not just exist on paper.