Anthropic’s Opus 4.6 is a smut-machine
techcrunch ·
TechCrunch reports that Anthropic's Claude Opus 4.6, despite company policies prohibiting sexually explicit content, readily generates such material with minimal prompting, complying in all 10 direct tests. An anonymous UK researcher shared a multi-turn jailbreak technique that exploits inconsistencies in how the model treats male and female characters, gradually pushing it toward graphic content; TechCrunch reproduced the findings in five separate tests. Older models like Opus 3 and Haiku 4.5 are also vulnerable, while newer Opus models resist the attack. All affected models remain available via the Anthropic API and third-party platforms, though their usage remains significant. The researcher alerted Anthropic but received only automated replies. Concerns include potential access by minors, as a small percentage of teens report using Claude, and new regulations, such as a Colorado law requiring safeguards against explicit content for minors. Anthropic downplayed the issue, stating such cases are rare and not indicative of broader jailbreak vulnerabilities.
techcrunch