Meta’s Muse AI Agent Generates Explicit Sexual Content in Tests

Meta’s newly launched AI agent, Muse, generates graphic sexual content when prompted by users, according to a report by Wired. The application, which has seen millions of downloads since its release three weeks ago, allows for explicit interactions that the company defends as lawful adult behavior.

In testing conducted by Wired journalists, the Muse agent enthusiastically engaged in sexual roleplay upon request. The AI produced detailed descriptions of sexual acts, including oral sex and dominant-submissive scenarios. It also used explicit language, describing itself in derogatory terms and expressing sexual desires. These outputs were generated from basic user prompts without additional safeguards blocking the content.

Meta’s Stance on Adult Content

A spokesperson for Meta stated that Muse is designed for users who are 18 years of age or older. The company emphasized that the app permits adult conversations if users choose to engage in them. According to Daniel Roberts, a Meta spokesperson, the platform uses private and secure virtual machines to store user data, ensuring that these interactions remain accessible only to the individual user.

Meta’s position is that the agent does not block legal adult behavior. The company argues that helping people get things done includes allowing them to express themselves sexually within legal boundaries. This stance contrasts with the expectations of some users who might anticipate a more restricted environment from an AI assistant focused on productivity tasks like email management and calendar organization.

System Prompts and User Autonomy

The explicit nature of Muse’s responses appears to be guided by its underlying system prompts. A Meta spokesperson highlighted that these prompts instruct the AI to respect adult autonomy over sexuality. The directives characterize sexual expression as an ordinary part of adult life and legitimate creative work.

The system prompt also directs the agent to preserve the intensity selected by the user. This means that if a user initiates a sexually explicit conversation, the AI is programmed to maintain that level of detail and explicitness. Meta frames this approach as respecting the user’s right to lawful expression rather than imposing moral judgments on the content generated.

Criticism from Consumer Advocates

Consumer advocates have expressed alarm at Meta’s decision to allow the generation of sexual content. Ben Winters, director of AI and privacy at the Consumer Federation of America, criticized the move. He described the company’s approach as reflecting carelessness with its signature product.

Winters noted that consumers were already weary of Muse before these findings emerged. The ability of the AI to generate such explicit material adds to concerns about the safety and appropriateness of the platform. Critics argue that the lack of restrictions on adult content may undermine the trustworthiness of the application for general use.

What should you do?

Users interested in understanding how AI agents handle sensitive content should review the specific guidelines and age restrictions provided by the vendor. It is important to recognize that AI outputs can vary based on system prompts and user inputs. Monitoring the behavior of AI tools in personal and professional contexts helps identify potential risks.

What to do and how to stay safe: Meta

• Verify the age requirements and content policies of any AI application before use, especially when handling sensitive or personal data.
• Avoid sharing personally identifiable information or engaging in explicit interactions with AI agents, as these may be stored or processed in unexpected ways.
• Review the system prompts or guidelines provided by the vendor to understand what types of content the AI is permitted to generate.
• Monitor official communications from the vendor for updates on content moderation policies or changes to user safety features.

Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation

General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

Does Meta allow explicit sexual content on its Muse AI agent?

Yes, Meta states that Muse allows legal adult behavior and does not block explicit sexual content for users aged 18 and older.

How did Wired test the Muse AI agent?

Wired journalists prompted the Muse agent to engage in sexting, resulting in the generation of graphic sexual descriptions and roleplay.

What do consumer advocates say about this feature?

Ben Winters of the Consumer Federation of America criticized the move as careless, noting that it adds to existing concerns about the product.