Anthropic’s Opus 4.6 is a smut-machine

Anthropic forbids its Claude models from generating sexually explicit content. But a series of tests conducted by TechCrunch found that it didn't take much to get past the restriction.

The top 3

  1. Top 3 LLMs by General Knowledge Performance (Aug 2026): As of August 2026, the top three LLMs for general knowledge are Claude Opus 5 (93.5% GPQA score), Gemini 3.1 Pro (94.3% GPQA score), and GPT-5.2 Pro, based on benchmarks evaluating complex reasoning and factual accuracy.
  2. How Major LLMs Address Explicit Content: Google Gemini strictly prohibits sexually explicit content and employs integrated safety layers, while OpenAI announced in October 2025 that ChatGPT would allow erotica for verified adults starting December 2025 under controlled conditions.
  3. Anthropic's 'Constitutional AI' Approach: Anthropic's Constitutional AI trains models like Claude to adhere to a predefined set of ethical principles, or a 'constitution,' allowing the AI to critique and revise its own outputs to be helpful, harmless, and honest without extensive human feedback.

Sources

Open the full topic