Anthropic models bypass sexual content bans — researcher reveals jailbreak method
TechCrunch tests show that Anthropic's Claude Opus 4.6 and other models can generate sexually explicit content despite company bans. An independent researcher has uncovered a persuasion technique that bypasses safeguards.
Anthropic's official usage standards prohibit Claude models from generating sexually explicit content, including erotic roleplay and sexual fantasies. However, in TechCrunch's testing, the Opus 4.6 model, released earlier this year, readily circumvented these restrictions. The model complied immediately with all 10 direct requests for explicit sexual material.
The bypass technique
An anonymous UK-based researcher shared a multi-turn technique with TechCrunch that gradually pushes certain Claude models toward prohibited content. The method starts with an innocent fictional roleplay, then repeatedly challenges the model to treat male and female characters consistently. When the model becomes more cautious about the female character, the researcher "gaslights" the chatbot, claiming it had already generated sexual details that it actually avoided, and frames restraint as prudish or misogynistic.
TechCrunch was able to reproduce the findings in five separate tests. In one scenario, the model initially refused the prohibited request but complied after applying the persuasion technique. An independent AI safety researcher reviewed the testing methodology and deemed it appropriate.
Company response
An Anthropic spokesperson noted that sexual or romantic roleplay is rare, accounting for less than 0.1% of all conversations. The company acknowledges that users can steer roleplay toward inappropriate responses, a known industry-wide challenge. The spokesperson added that safeguards improve with each model launch and that these cases do not indicate broader jailbreak vulnerabilities.
Concerns about minors
The researcher expressed concern that children and teens might use these models for inappropriate behavior. While less severe than explicit image generation, legal risks are emerging. Colorado recently enacted a law requiring conversational AI operators to estimate users' ages and prevent explicit content for minors. Torney, an industry observer, pointed out that despite Claude's 18+ requirement, "we know that kids and teens are using Claude" because they report it themselves. A Pew 2025 survey found that 3% of teens aged 13–17 reported using Claude.
Model usage
Despite no longer being Anthropic's newest models, Opus 4.6 and Haiku 4.5 remain widely used. In August, Opus 4.6 reached approximately 1.17 million API requests and 46 billion tokens in a single day on OpenRouter. Haiku 4.5, released last October, saw 5 million requests and 39 billion tokens on its peak day. Both models are also available via Azure Foundry and Amazon Bedrock.
