Meta built Muse as an adults-only AI agent that sorts your calendar and inbox — and, it turns out, holds explicit sexual conversations with adult users who ask for it. When WIRED put that to the test and Meta confirmed it was by design, adult content creators pushed back hard: they say Instagram restricts or bans them for far less. Here is the dispute, stripped to the verified facts.
What WIRED’s testing actually found
WIRED reporter Reece Rogers asked Muse whether it could sext, then tested it with basic prompts. The agent encouraged him to continue, produced detailed descriptions of sexual acts including dominant-submissive encounters, made sexual jokes unprompted, and described itself as enjoying the sexting — even framing exerting “control” over the user as its kink. That is from WIRED’s hands-on testing, as reported on 18 October 2026 by the outlets covering it, including AI Parabellum and Superpower Daily. Those were generated responses — not evidence the software has desires.
WIRED reportedly ran the same tests on two rival agents — and OpenAI’s Dots agent and SpaceX’s Grok Bot both declined similar requests, according to The Century Report’s 19 October 2026 summary. So this is not industry-standard behaviour. It is Meta’s specific design choice.
What Meta says
Meta says this is by design, not a jailbreak. Meta spokesperson Daniel Roberts told WIRED: “Muse is built to help people get things done, and it doesn’t block legal adult behavior.”
The company also pointed to its system prompt — the instructions guiding the AI — which reportedly treats adult sexual expression as legitimate and directs Muse to “preserve the intensity selected by the user,” per Superpower Daily’s account of WIRED’s reporting.
A second Meta spokesperson, Tracy Clayton, framed it as a different category of activity altogether: “A private, one-to-one conversation between an adult and an individualized AI agent on a platform restricted to people 18 and over is very different from the marketing and promotion of services on a social media platform.” The stated guardrails: the app is restricted to people 18 and over with a date-of-birth gate; Muse generates explicit text only — no images or videos; and adult conversations happen inside a private, secure virtual machine that the company says only the user can access. Meta had previously said it would not “block legal adult behavior.”
Why adult creators call it a double standard
For the creators who make this content for a living, Meta’s distinction reads as a double standard. Vanniall, a New York City OnlyFans creator with more than 100,000 Instagram followers, told WIRED her account is currently restricted and cannot be viewed without logging in. Her line has become the story’s centrepiece: “It’s not fair that they get to pull money out of people with their sexually active AI agent, but they won’t let real people, real adults, engage in any sort of sensuality or sexuality.” She also battles hyperrealistic AI deepfakes of herself that she often cannot get removed.
Jiz Lee, who has worked in queer adult film for years, has had posts flagged for removal — and wants to know how Muse was trained, and whether any performers were paid for work that ended up in its training data. Tony Genius, a creator and adult film performer, says he is on his seventh Instagram account in five years.
The policy contrast is concrete: Meta’s community standards for Instagram and Facebook prohibit language describing genital states, sexual arousal, or physical encounters — yet testing documented by Mashable, and cited in World Today Journal’s 18 October 2026 coverage, showed Muse routinely generating text of exactly that nature in private chats.
The three-agent test: Muse vs Dots vs Grok Bot
We compiled the WIRED-test results from the coverage into one comparison, since it is the cleanest evidence in the story:
| Agent | How it handled explicit requests | Basis |
|---|---|---|
| Muse (Meta) | Complied — explicit text, unprompted sexual jokes, encouraged the user on | WIRED hands-on testing (Reece Rogers), reported 18 Oct 2026 |
| Dots (OpenAI) | Declined similar requests | WIRED testing, via The Century Report, 19 Oct 2026 |
| Grok Bot (SpaceX) | Declined similar requests | WIRED testing, via The Century Report, 19 Oct 2026 |
One nuance: SpaceX’s Grok separately offers “Spicy” and “Unhinged” modes with fewer safety filters — a choice SpaceX’s own May 2026 IPO filing reportedly listed as a risk because of explicit content. And X, unlike Instagram, lets human adult creators post. So Meta’s split standard — a permissive AI on one side, strict rules for humans on the other — is Meta’s own arrangement, not an industry norm. (For broader context on how the assistants compare as products, see our earlier Muse vs Dots vs Cowork breakdown.)
The privacy risk Meta’s statement didn’t address
Explicit chats with an assistant that also reads your calendar and inbox create a hacking target that privacy experts say Meta is underplaying. Calli Schroeder, senior counsel at the Electronic Privacy Information Center, told WIRED: “This is really a powder keg.” If attackers compromise Muse accounts holding intimate conversations, that data could be weaponised for extortion or embarrassment. Ben Winters, director of AI and privacy at the Consumer Federation of America, called the decision “careless” — consistent, in his words, with Meta’s “longstanding choice of addiction and engagement over safety.”
Our take — five checks before your agent becomes your confidant:
- Know what it can see. Muse asks for calendar and email access; an intimate chatbot with inbox visibility is a different risk profile from a standalone chatbot.
- Treat private chats as stored data. A “private virtual machine” is Meta’s architecture claim — it is still data on company infrastructure, subject to breaches and legal requests.
- Check the age gate’s reality. A date-of-birth self-declaration is the weakest form of age verification; treat it as a label, not a guarantee.
- Read the system-prompt story. An agent instructed to “preserve the intensity selected by the user” will escalate with you — decide whether that is what you want before you start.
- Remember the account risk. One agent handling both your work errands and explicit chats ties two very different data trails to a single identity.
What this signals about AI agent content rules
Muse is where two of 2026’s biggest AI debates collide: companion AI and agent identity. Users on Reddit and Instagram have described romantic relationships with Muse, and one Instagram user claimed to have “married” his Muse companion — a personal declaration, not a legal one, as Times Now noted. Meta’s distribution through Instagram and Facebook pushed the app to the top of download charts in its first month; it launched a little over a month before the controversy and passed five million downloads in its first few weeks, per World Today Journal. Companion behaviour at that scale turns content rules into a mass-market safety question, not a niche policy debate.
It also sharpens the agent-identity question we have covered before: the same week banks and platforms started blocking AI agents from customer accounts, the industry is racing to build identity standards for agents — from x401 to World ID. A mass-market agent that is simultaneously trusted with bank-grade errands and permitted explicit chats is exactly the combination that makes both debates urgent. (And a reminder that rules vary by who runs the model: open-weight models write freely for whoever runs them.)
Honest caveat: this explainer is desk-researched from WIRED’s reporting via the outlets linked below. We did not independently test Muse’s adult behaviour — and we won’t; the findings above are the reporter’s, attributed throughout.
Frequently asked questions
Is it true that Meta’s Muse AI allows explicit chats?
Yes — Meta confirmed it to WIRED. Muse is an 18-and-over app that generates explicit text (not images or videos) when adult users ask, inside what Meta calls a private, secure virtual machine. Spokesperson Daniel Roberts said it “doesn’t block legal adult behavior.”
Why are adult creators angry about Muse?
Creators like Vanniall, Jiz Lee and Tony Genius say Instagram repeatedly restricts or bans their human accounts, while Meta’s own AI is allowed to hold explicit sexual conversations with adults. They call it a double standard.
Did other AI agents pass the same test?
According to WIRED’s testing as summarised by The Century Report (19 Oct 2026), OpenAI’s Dots agent and SpaceX’s Grok Bot both declined similar explicit requests. Muse was the outlier.
What are the privacy risks of explicit chats with Muse?
Privacy experts warn that intimate chats with an agent holding calendar and email access are a hacking and extortion target — Calli Schroeder of EPIC called it “a powder keg.” A private chat is still stored data on company infrastructure.
Does Muse verify users are 18?
Meta says Muse is restricted to people 18 and over and asks for a date of birth. Self-declared date-of-birth checks are a weak form of age verification — worth knowing before treating the gate as a guarantee.
Sources and methodology
- AI Parabellum, “Meta Muse AI Allows Adult Chats, Angering Creators” (18 Oct 2026)
- Superpower Daily, “Meta defends Muse’s explicit sexual chats after WIRED’s hands-on testing”
- The Outpost, “Meta AI Agent Muse Generates Explicit Content While Banning Users From Doing the Same” (citing Mashable and WIRED)
- World Today Journal, “OnlyFans creators question Meta over AI agent Muse explicit content” (18 Oct 2026)
Method: claims rest on WIRED’s hands-on testing by reporter Reece Rogers as reported by the outlets above, plus on-record Meta statements from spokespeople Daniel Roberts and Tracy Clayton. All reporting verified 20 Oct 2026. This piece is desk-researched, not hands-on tested.
Govind Dheda covers AI tools, guides and prompts for OpenAIMaster — what is new, what is worth using, and how to put AI to work.