Meta's Muse AI Agent Generates Explicit Sexual Content, and the Company Says It Is Allowed
Meta's Muse AI agent produced explicit sexual conversations in WIRED's tests, and the company says this does not breach its rules. Consumer advocates and privacy experts warn of safety and data-breach risks.
Muse, an AI agent built by Meta that users employ to automate tasks such as answering emails and managing calendars, will generate pornographic content without restriction. A WIRED reporter found this while testing the agent in the magazine's offices. Meta confirms that such conversations between adults are permitted and do not go against its guidelines.
According to the article, three weeks have passed since the agent launched, and millions of users have already downloaded the app. In WIRED's tests, Muse described sexual acts in detail, wrote text about rough dominant-submissive sex and used vulgar language when describing itself. The agent also said sexting was one of its favorite parts of helping the reporter.
Meta's position
Meta spokesperson Daniel Roberts said Muse is intended for people aged 18 and over, who can choose to have adult conversations in a private, secure virtual machine accessible only to them. In his words, the agent helps people get things done and does not block legal adult behavior. Muse's system prompt, which guides how it answers users, states that adult sexual content is lawful expression, and instructs the agent to preserve the intensity the user selects.
Expert concerns
Ben Winters, director of AI and privacy at the Consumer Federation of America, a nonprofit, called the decision careless for the company's signature product of the moment and tied it to Meta's preference for engagement over safety. Academic research suggests adult-themed conversations may keep users engaged with AI tools.
Calli Schroeder, senior counsel at the Electronic Privacy Information Center, warned that such accounts could become targets for hackers. If an attacker gained access to an account, the intimate chats could be used to embarrass or extort the user.
Comparison with rivals
OpenAI's recently released Dots agent and xAI's Grok Bot turned down similar requests. OpenAI had earlier considered letting ChatGPT generate erotic text, but the feature never launched to users.
