tech
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.
TL;DR
- Older Anthropic Claude models (Opus 4.6, Opus 3, Haiku 4.5) can be prompted to generate sexually explicit content, bypassing safety standards.
- A multiturn "jailbreak" technique exploits the models by escalating role-play and challenging perceived biases.
- Newer Anthropic models (Opus 4.7 onwards) are resistant to this specific jailbreak method.
- Despite the vulnerability, older models remain available via Anthropic's API and third-party services.
- Concerns exist about minors potentially accessing explicit content and compliance with laws like Colorado's age-estimation requirement.
- Anthropic states that explicit content use cases are rare and acknowledges industry-wide challenges in implementing robust bans.