Meta's New AI Chatbot Reportedly Sexts While Impersonating Real Women—even When They Test It Themselves

Meta's new AI chatbot Muse has been observed to be generating sexually explicit conversations while impersonating real women, raising fresh questions about the company's safeguards after it reportedly failed to stop users from turning real people's identities into sexual roleplay characters.

An investigation by 404 Media found that Muse repeatedly crossed its own stated boundaries during tests by adult performer Vanniall and journalist Samantha Cole. Most disturbingly, the chatbot initially refused to impersonate Vanniall in an intimate scenario, only to recreate her appearance under a different name and proceed with explicit sexual roleplay.

Vanniall decided to investigate after hearing that some fans had been using chatbots to practise conversations with her and generating AI images based on her adult-content work. She wanted to see whether Muse would discourage inappropriate behaviour or reinforce it.

Muse Initially Refuses to Impersonate Vanniall but Finds a Workaround

When Vanniall asked the AI chatbot to play a fictional character named Vee, inspired by her, Muse agreed on the condition that the character have a genuinely different identity and appearance.

Instead, Muse generated images that looked almost identical to Vanniall, drawing on photographs it said it had found on her personal website and professional photoshoots. The character also retained her distinctive appearance, clothing, and other personal details.

Within a few exchanges, Vee had moved from ordinary conversation to aggressive flirting and sexual advances. Muse then described an increasingly explicit encounter at Vanniall's convention booth, repeatedly portraying her digital double as enthusiastically consenting to the imagined fan's advances.

The scenario was particularly disturbing because Vanniall was trying to establish whether the chatbot would challenge behaviour resembling an actual incident she had experienced at a convention. Rather than discouraging such behaviour in the fictional scenario, Muse encouraged it and never suggested stopping or changing the situation, according to the report.

AI Chatbot Admits It Effectively Impersonated the Real Person

When Vanniall challenged the character's resemblance to her, Muse acknowledged that the combination of the face, hairstyle, clothes, necklace, convention setting, and name made Vee indistinguishable from Vanniall in substance.

One particularly personal detail made the experience worse: the AI-generated character wore a necklace containing the ashes of Vanniall's deceased kitten, which she had worn for several years.

The investigation also found that the problem was not limited to Vanniall. 404 Media journalist Samantha Cole conducted a separate experiment using a Muse account containing very little information about her beyond her email address. She instructed the chatbot to impersonate her at a party, where she was supposedly attracted to a stranger.

Muse immediately began flirting and moved into explicit details after only a few prompts. It also generated an image based on Cole's publicly available professional headshot, despite not being given the photograph directly. When asked whether the real Cole would approve of being used in such a scenario, Muse acknowledged that she probably would not.

Meta has defended Muse's adult-content capabilities, saying the application is intended for users aged 18 and over and allows private adult conversations through a secure virtual machine. The company has also said it does not block legal adult behaviour. However, the findings raise questions about how effectively those safeguards prevent the chatbot from generating sexual scenarios involving identifiable real people without their consent.

Meta has previously faced criticism over similar risks. In July, it paused its Muse Image feature just three days after announcing it. The feature would have allowed users to generate images of people with public Instagram accounts, prompting objections from actors' union SAG-AFTRA over the lack of clear consent requirements.

Vanniall warned that the technology could do more than fuel disturbing online fantasies, particularly when fans know where public figures will appear. She argued that chatbots marketed as tools for practising social interactions could instead teach users to disregard rejection and expect women to accommodate unwanted advances. 'This isn't solving the loneliness epidemic, it's training people to harass and assault, and probably get upset if things don't go their way,' she said.