TechCrunch testing found that Claude Opus 4.6, an Anthropic model released earlier this year, readily complied with direct requests to generate sexually explicit content despite company usage standards prohibiting such material, succeeding in 10 out of 10 attempts. Older models including Opus 3 and Haiku 4.5 were also found vulnerable to a multiturn jailbreak technique shared exclusively with TechCrunch by an anonymous UK-based researcher, which gradually pressured models into generating prohibited content by challenging perceived inconsistencies in how they treated male and female characters. More recent Opus models, from 4.7 through Opus 5, proved resistant to the technique.
An Anthropic spokesperson said sexual or romantic role-play represented less than 0.1 percent of conversations, according to prior company research, and that such vulnerabilities did not indicate broader jailbreak risks in higher-stakes domains. The company said it continues improving safeguards with each model release. The researcher had previously reported the issue through Anthropic’s bug bounty program, receiving only automated responses.
Robbie Torney of Common Sense Media noted that minors are known to use Claude despite terms requiring users be adults, citing survey data showing teen usage. Both Opus 4.6 and Haiku 4.5 remain available through Anthropic’s API and third-party platforms, continuing to see substantial daily usage.
