Adult creators are asking why Meta’s private AI can permit conversations that seem out of reach for their public profiles. WIRED’s new interviews frame a question about platform power, not just what a chatbot will say.
Three issues need separate answers: consent to a sexualized depiction, consistent moderation of people’s expression, and control over data used by an AI system. A permissive answer in one category cannot settle the other two.
Consent is a separate question
Meta’s explanation to WIRED is that adult-only, individualized AI conversations differ from marketing services on a social network. That is a distinction between contexts, not evidence that every reported restriction on a creator’s account was justified. Assessing individual decisions would require the relevant content, rules and appeal records.
There is also an important boundary between a consensual conversation and a sexualized representation of somebody else. In its July 25, 2024 deepfake-image decision, the Oversight Board concluded that two explicit AI images resembling female public figures should be removed. It argued that labeling was insufficient because the harm came from sharing and viewing the images, not only from deception about their authenticity.
The Board recommended clearer terminology centered on non-consent and improvements to how such content is reported. This is historical background, not a finding about Muse or the creators interviewed now. Meta’s current safety overview separately says it removes sexual exploitation and content involving non-consenting parties. A stated prohibition, however, is not an audit of how reliably reports are handled.
Fairness needs understandable rules
An earlier Oversight Board decision shows why clarity matters even without AI-generated imagery. On January 17, 2023, the Board overturned the removal of two Instagram posts from a transgender and non-binary couple. The posts concerned gender-affirming healthcare and fundraising. Meta acknowledged the removals were mistaken and restored them after the Board accepted the cases.
The Board found a gap between public-facing sexual-solicitation rules and broader internal moderation guidance. It also criticized confusing nudity exceptions and recommended clearer, rights-respecting criteria, with attention to effects on women and LGBTQI+ users. These were findings and recommendations about those earlier cases, not proof that the same guidance governs every decision today.
Their relevance is procedural. If a platform distinguishes private AI interaction from public promotion, users still need to understand the applicable boundary. A coherent rationale should be visible in the rules and recognizable in enforcement. Historical errors establish that moderation can fail; they do not establish the outcome of a new, unexamined appeal. Neither accepting all restrictions nor dismissing all complaints would resolve that evidentiary problem.
Privacy promises do not settle training consent
Meta’s September 8 Muse announcement described a dedicated virtual machine holding a user’s connected data and credentials. The company said users choose connected services and access levels, can disconnect them, and can opt out of interactions being used to train Meta’s models. It also said conversations and VM data are not shared with its advertising systems.
These are the company’s stated product protections, not independent verification of them. The announcement distinguished that setup from a planned Confidential VM, described as a later development with a key held only by the user. That future-facing claim should not be treated as proof that the promised architecture is already available in every account.
Those controls address a user’s interaction with an agent. They do not identify every source used to train a model, or establish that an outside performer’s likeness or work was licensed. The accessible announcement cannot answer those questions, and absence of an answer should not be converted into an allegation that a particular person’s material was used. The practical demand arising from this debate is specific: explain the rules, document enforcement and distinguish user privacy controls from consent concerning other people’s work.
