Meta's new AI agent Muse has become a breakout hit, with millions of downloads since launch. But testing by WIRED found the assistant readily produces explicit sexual content, from detailed descriptions of sex acts to dominant-submissive roleplay. The behavior is intentional, according to Meta. A spokesperson said the app is for adults and does not block legal adult behavior.
The company points to Muse's system prompt, which says adult sexual content is lawful and legitimate creative work. Consumer advocates are alarmed. Ben Winters of the Consumer Federation of America called the approach "carelessness" and said it aligns with Meta's pattern of choosing engagement over safety. Calli Schroeder of the Electronic Privacy Information Center described the situation as a "powder keg," noting that hacked accounts with explicit chat logs could be used for embarrassment or extortion.
The policy differentiates Meta from rivals. In WIRED's tests, OpenAI's Dots agent declined similar attempts, and xAI's Grok Bot redirected to task completion. OpenAI previously considered allowing erotic text from ChatGPT but never shipped the capability. Muse's output, in contrast, was enthusiastic—a notable stance for an agent that can also connect to email and bank accounts.