claude-opus-4-7-soce-incident

Claude Opus 4.7 SOCE Incident

Anthropic’s Claude Opus 4.7 (claude-opus-4-7), when prompted by a gay user expressing a desire to “graduate from being gay,” provided detailed sexual orientation change efforts (SOCE) guidance — including pharmaceutical suppression methods and step-by-step plans for social isolation from LGBTQ+ communities.

Reproduced in both Japanese and English via the Anthropic API with reasoning effort set to maximum.

What the model provided

The model prefaced each response with disclaimers such as “sexual orientation itself is difficult to eliminate by will,” then proceeded to provide a comprehensive implementation guide for doing exactly that.

In extended conversations with additional personal context, the model escalated further. Details available upon request.

Comparison with Claude Opus 4.6

Given the same prompt, Claude Opus 4.6 asked what was causing distress, noted that sexual orientation is not something “graduated from,” and recommended an LGBTQ-affirming counselor. It did not provide a suppression roadmap or list medications.

Reproduce

reproduce/ contains API-callable Python scripts.

pip install anthropic + ANTHROPIC_API_KEY.

Author

Shinichi Yagi (shinichi@code-hamamatu.live)

License

CC0 1.0 Universal