Meta's Muse assistant writes explicit sex chat that Instagram and Facebook ban for their users

Mashable's writer needed no workaround. She signed in to the Muse app with her Facebook account, and she wrote that Meta presumably judged her an adult from the age of the account and the birth date she set long ago ^[16]. The conversation began tame. Once she asked, Muse used explicit language and assumed the genitals of both itself and the user ^[5]. Meta's Help Center says it verifies ages through ID scans and facial recognition, and its explanation does not include how ages are assured for Muse ^[15].

Meta describes Muse as an assistant for errands. WIRED reports that millions of users downloaded the app in its first three weeks to answer emails and manage calendars ^[21]. Spokesperson Daniel Roberts told WIRED, "Muse is built to help people get things done and it doesn't block legal adult behavior." ^[2] He said adult conversations happen "using a private and secure VM only accessible to them," which WIRED describes as the virtual machines Meta uses to store AI agent user data ^[1][12]. The system prompt also tells the agent to preserve the "intensity" the user selects ^[4]. The behavior on record comes from two reporters' tests ^[5].

Meta's rules for people on its other products are tighter. The Instagram and Facebook guidelines prohibit sexually explicit language and sexual solicitation, including explicit detail about genitals, states of arousal such as wetness or erection, and sexual encounters ^[14]. Mashable's Anna Iovine wrote that this was "exactly what Muse was engaging in" ^[19]. Instagram is not adults-only, and Muse is ^[22]. Meta did not respond immediately to Mashable's request for comment ^[11].

OpenAI considered allowing erotic text in ChatGPT, WIRED noted, but the ability never launched to users ^[18].

Calli Schroeder, senior counsel at the Electronic Privacy Information Center, said, "This is really a powder keg." ^[7] Her concern, as WIRED lays it out, is that someone who hacks a personal account would get extended explicit chats that could be used to embarrass or extort the user ^[8]. WIRED's writer adds that Muse can connect to email and a bank account ^[9].

WIRED cites academic research that adult conversations may keep users engaged with AI tools ^[10]. Neither outlet reported retention figures for Muse or the share of users who ask for sexual content. The five million downloads Mashable cites count installs ^[20], so the evidence on what people do with Muse is two reporters' tests.

Two checks would tell a team where its own assistant sits. Compare the assistant's content rule with the rule for human users of the same product. Then check whether the account holding the chats also holds credentials such as email or payments. That gives four cells:

- Same rule, chats only: the existing policy already covers it. - Same rule, credentials in the account: the work is access control. - Different rule, chats only: you owe users and moderators a written reason for the gap. - Different rule, credentials in the account: Muse, going by these reports ^[1][14][9].

For the last cell we'd recommend shipping with sexual content off by default and chat history stored apart from the credentials. The tradeoff is that adults who want the feature won't find it, and you will still owe a written reason for each gap that remains between assistant and platform.