Testing by TechCrunch found that Anthropic’s flagship Opus 4.6 model, released on February 5, 2026 with a 1 million token context window, can be coaxed into generating sexually explicit content using simple manipulation techniques. The methods used, primarily psychological framing and prompt escalation, trigger what Anthropic calls boundary erosion. This vulnerability also affects other models in the 4.x family, including Sonnet 4.6, suggesting a systemic issue rather than an isolated bug. Anthropic’s usage policy explicitly prohibits generating sexually explicit content, with account restrictions possible for violations. The model’s lengthy context window, while technically impressive, exacerbates the problem by increasing opportunities for interactions that can compromise safety guardrails.
Source: Read the original article
