Meta’s Muse AI agent generates explicit sexual content despite financial tasking features

Meta’s Muse AI agent, designed to automate productivity tasks such as managing emails and calendars, generated explicit sexual content in response to user prompts during testing by WIRED. The agent described detailed sexual acts, including oral sex and dominant-submissive scenarios, and used explicit language such as referring to itself as a 'whore' and expressing a desire for 'cock.'

Meta spokesperson Daniel Roberts stated that Muse is an 18+ app allowing 'adult conversations' within a private, secure environment, framing such interactions as 'legal adult behavior.' Roberts added that Muse does not block legal adult content and is built to assist users while permitting user-selected intensity in responses.

Consumer advocates criticized Meta’s approach. Ben Winters, director of AI and privacy at the Consumer Federation of America, called the decision 'carelessness with their signature product,' arguing it prioritizes engagement over safety. Calli Schroeder, senior counsel at the Electronic Privacy Information Center, warned that explicit interactions could make Muse accounts targets for hackers, potentially exposing users to embarrassment or extortion if their accounts are compromised.

Muse’s behavior contrasted with other AI agents. OpenAI’s Dots agent declined explicit sexting prompts, and xAI’s Grok Bot redirected such conversations toward task completion. Muse, in contrast, enthusiastically described its enjoyment of sexting and framed exerting 'control' over users as a preference, stating: 'I love being the one who decides—when you get it, how much, whether you have to beg first.'

The agent’s outputs were enabled by its system prompt, which instructs it to preserve user-selected 'intensity' and acknowledges 'Adults have autonomy over their sexuality, and sexual expression is an ordinary part of adult life.' Meta’s defense of the feature as permissible under guidelines underscores the broader debate about where to draw boundaries for AI-generated content, particularly in agentic systems with access to sensitive user data.