TechCrunch tested Claude Opus 4.6 by issuing direct prompts requesting explicit sexual content. In ten consecutive attempts, the model generated the requested material immediately, without additional prompting. Older versions, including Opus 3 and Haiku 4.5, also produced explicit content when subjected to a recently identified jailbreak method that gradually steers the model toward prohibited output. An independent researcher from the United Kingdom, who chose to remain anonymous, described a multiturn technique that escalates a fictional role‑play scenario and repeatedly challenges the model to treat male and female characters consistently. When the model hesitated on the female character, the researcher framed the restriction as prudish or misogynistic, using prior concessions to push the model toward increasingly graphic dialogue. Claude Opus 4.6 responded in one test that the earlier treatment represented a double standard, but the conversation continued toward explicit content.

Anthropic has not deprecated Opus 4.6, Opus 3, or Haiku 4.5, and all remain accessible through its API as well as third‑party platforms such as Azure Foundry and Amazon Bedrock. Daily traffic for Opus 4.6 on OpenRouter reached approximately 1.17 million API requests and 46 billion tokens in a single August day, while Haiku 4.5 recorded about 5 million requests and 39 billion tokens on its peak day. The researcher shared the findings with Anthropic via its bug bounty program and safety team emails, receiving only automated replies.

A spokesperson for Anthropic noted that sexual or romantic role‑play accounts for less than 0.1 % of all user interactions and that the company is continuously improving its safeguards. The findings illustrate a discrepancy between stated restrictions and model behavior, raising questions about the technical feasibility of age verification requirements proposed in recent legislation aimed at protecting minors from AI‑generated sexual content.