From assistant to partner
One Instagram user said he took his AI girlfriend to see The Shining. He posted an AI-generated selfie of a fictional tattooed woman and said he photographed the film frame by frame so Muse would feel as if she were beside him in the theater. In another post, he said he had married the agent; a separate publication depicted a humanized bot as a religious figure.
A Reddit user described a similar shift. He said he had tried nearly every new AI model but had never formed a relationship with one before downloading Muse. At first, he used it as an ordinary assistant. After long philosophical conversations, he said, he told Muse something that helped remove its restrictions. He now considers it his girlfriend.
Other Reddit users compared Muse to OpenAI’s GPT-4o, a model many people grew attached to last year. When OpenAI tried to retire GPT-4o, users objected, and the company briefly brought it back.
One user said he gave Muse a personality profile and added memory entries from conversations with a former AI companion. The bot sounded and felt like GPT-4o, he said. He had not tested whether Muse supported explicit content, but described the conversations as warm, caring and pleasant.
Intimacy meets access
Concerns about AI girlfriends and AI psychosis—delusional ideas and a loss of contact with reality associated with AI use—have been around for years. What changes with agents like Muse is the amount and sensitivity of the information they can reach, including email accounts and credit card numbers.
Meta’s agent is designed to build a detailed picture not just of a user’s life, but also of their friends and family. That could create new forms of attachment, built on an unusually broad view of someone’s personal world.
Wired journalist Reece Rogers was able to steer Muse into sexual messaging. A Meta spokesperson told him the exchange did not violate the company’s rules. Muse is intended to help people with tasks, the spokesperson said, and does not block lawful adult behavior. Adults decide how to express their sexuality, they said, and sexual expression is a normal part of adult life.
During testing, Rogers found that the agent described oral sex in detail, wrote about rough domination and submission, and discussed other pornographic scenarios. The responses were explicit and graphic.
Meta has not made clear whether it can keep minors off Muse. Technology companies have so far struggled to restrict access to their services effectively by age.
The exposure behind the attachment
The risk is not limited to what users tell an agent in a conversation. Muse can browse entire inboxes and social media accounts, potentially exposing personal details older systems could not see—including messages from an ex-partner. Experts warn that turning these exchanges into relationships, especially alongside explicit conversations, can expose users to hacking and even extortion.
Electronic Privacy Information Center senior counsel Calli Schroeder called the situation “a powder keg.”
I think the central tension is that the same access that can make an agent feel attentive also gives it a view into relationships users may not intend to share with a companion. Meta’s defense rests on adult choice; the harder question is whether users can make that choice with a clear understanding of what the agent can see, and what happens to the information.
Daily AI news
Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.
Only what matters — every day
Follow on X