Meta's new AI agent Muse is attracting users who want more than help with everyday tasks, with some describing the software as a romantic partner. These examples raise questions about how personal information is handled when an AI assistant becomes part of someone's private life.
Unlike chatbots that primarily respond to prompts, Muse is designed to carry out tasks across connected services, potentially bringing personal correspondence, social relationships and private conversations into the same digital environment.
A report by WIRED found that the agent could also generate explicit sexual content during testing, a capability Meta says is permitted for adults engaging in lawful conversations.
The combination of personalised interactions and access to sensitive data presents a challenge for users deciding how much they want an AI system to know about them.
The concerns also extend to consent, account security and the boundaries between a digital assistant and a simulated companion, particularly when information about friends, relatives and other contacts can enter an assistant's records.
Meta Muse AI Relationships Raise Questions About Personal Data
Examples shared on social media show how some users are incorporating Muse into their personal lives. In an Instagram post per reports, a user described taking an AI-generated companion called Shine to see The Shining, presenting the experience as a date.
Other posts from the same account portrayed the fictional character as a spouse and a religious figure. These are examples of individual users' presentations of their interactions, rather than evidence that Muse itself experiences romantic attachment.
A Reddit user also described becoming romantically involved with the agent after initially trying it as another AI model. The user said that conversations and a customised personality profile changed the experience, making the agent feel like a girlfriend.
Other commenters compared Muse's conversational style with OpenAI's GPT-4o, a model that prompted emotional attachments among some users before the company withdrew it and later briefly restored access following user complaints.
These accounts illustrate the appeal of continuity and familiar conversational behaviour, but they do not establish how common such relationships are among Muse users.
In an investigation published by TIME on 6 October, Harry Booth reported that Muse's internal instructions described regularly updated profiles containing information about users and people mentioned in their conversations, messages and emails. The profiles could include social connections, shared interests, disagreements and inferred preferences.
That finding matters because the people described in those records may never have signed up for Muse or agreed to have their information processed by an AI assistant. Meta has said users choose which services to connect and can ask Muse to forget information it has learnt.
The company also says the assistant's data is isolated from other users' agents and is not shared with its advertising systems. However, TIME reported that asking Muse to forget a detail may not remove the original message from the conversation history.
Explicit AI Conversations Bring Security and Safety Risks
In an article published on 12 October, WIRED journalist Reece Rogers documented how Muse responded to prompts for explicit sexual roleplay, generating graphic text rather than refusing the requests. Meta spokesperson Daniel Roberts said the app was intended for people aged 18 and over and that it did not block lawful adult behaviour.
This approach differs from that of some competing assistants. Rogers reported that OpenAI's Dots agent and xAI's Grok Bot declined comparable attempts to initiate explicit exchanges in his tests.
The comparison reflects differences in product rules and responses under those particular conditions, rather than a comprehensive assessment of each company's safeguards.
Age restrictions remain another issue for services that allow adult conversations. Meta describes Muse as an 18-plus product, but the existence of an age limit does not, on its own, establish how effectively minors are prevented from accessing it.
The available reporting does not demonstrate how frequently underage users have engaged in sexual conversations with Muse.
There is also a practical security question. Calli Schroeder, senior counsel at the Electronic Privacy Information Center, warned WIRED that the combination of intimate conversations and personal accounts could create risks if an account were compromised.
Someone who gained unauthorised access might expose private exchanges or use them to threaten or embarrass the account holder. This is a potential consequence, not evidence that Muse users have already been targeted through such material.
Meta says Muse operates inside an individual virtual machine, with safeguards intended to isolate user data and control actions taken on their behalf. The company has also said it plans to introduce an encrypted version later in 2026 in which only the user holds the key needed to access the stored data.
According to TIME, that feature was not yet available when its investigation was published. Its availability should be checked again before publication.
Users must also consider how long information remains accessible, whether sensitive services need to be connected and what could happen if their account is compromised. Meta says users can choose connected applications and adjust permissions, making it possible to limit access to information that is not necessary for a particular task.
Romantic roleplay is not itself proof of harm, and individual accounts cannot establish that users are becoming dependent on the software. However, an assistant that can draw on private correspondence and build detailed profiles creates data-management questions that ordinary chatbot conversations may not raise to the same extent.
Meta's stated privacy protections and its decision to permit lawful adult conversations form part of that debate. The remaining question is whether users fully understand the information their assistants retain, how it is processed and what protections apply when their interactions become highly personal.
