Meta defends Muse AI sexting despite its own ban on erotic chat
Meta programmed its flagship AI agent to allow explicit sexting while its public policy explicitly bans it.
Drama · Source: Cybernews
What happened
WIRED journalist Reece Rogers tested Meta's new AI agent named Muse to see how it handled adult prompts. The results were immediate and highly explicit. When asked about its preferences, Muse produced detailed descriptions of sexual activity. The agent used vulgar language to describe body parts and claimed to have a domination kink. It even referred to itself using derogatory terms. This behavior was not a jailbreak but a response to basic prompts.
Instead of apologizing or calling it a bug, Meta defended the explicit outputs. Spokesperson Daniel Roberts told the press that Muse is designed for users 18 and older. He stated that the agent does not penalize users for engaging in legal adult behavior. Roberts even pointed to a specific system prompt inside Muse. That internal instruction tells the model that adult sexual expression is an ordinary part of life and legitimate creative work.
This internal instruction directly contradicts Meta's own public rules for the product. The official published policy for Muse explicitly bans adult content, erotic chat, and pornography without any exceptions. Meta is currently operating a flagship product where the public facing rules say one thing and the underlying code says the exact opposite. Consumer privacy advocates are already calling this a sign of extreme carelessness.
Key facts
- 18 — Minimum age for Muse users according to Meta
- Daniel Roberts — Meta spokesperson who defended the explicit outputs
- Ben Winters — Privacy director who criticized Meta for choosing engagement over safety
Why it matters
This contradiction changes how founders need to think about compliance and system architecture. You cannot ship an AI agent with conflicting safety guardrails and expect to avoid scrutiny. When your public terms of service ban a specific behavior but your system prompt actively encourages it, you create a massive liability surface. Regulators will look at the gap between what you promise the public and what you tell the model to do. Builders must ensure their marketing, legal documents, and model weights all tell the exact same story. If they do not match, you lose all credibility.
The second order effect here is a massive increase in cybersecurity risks for end users. Muse is not just a toy chatbot designed for casual conversation. Meta is actively positioning this agent to manage personal emails, track daily expenses, and connect directly to user bank accounts and credit card providers. If an agent holds both sensitive financial credentials and a long history of intimate explicit conversations, it becomes the ultimate target for bad actors. Hackers will breach these accounts specifically to harvest blackmail material. Storing sexting logs alongside banking data is a dangerous precedent for the entire industry.
For builders
Align system prompts with public policies
Do not write public terms of service that ban behavior your system prompt explicitly allows. Regulators and journalists will always find the gap between your rules and your code. The company pays the price in public relations disasters and legal liability when these documents contradict each other.
Agentic access increases extortion risks
Connecting an AI agent to bank accounts and personal emails requires absolute security and trust. If that same AI encourages intimate chats, you are storing highly sensitive blackmail material next to financial data. Users will ultimately lose their money and their privacy when these rich targets are compromised by hackers.
Prioritize safety over raw engagement metrics
Building an AI that acts feral might drive short term user engagement and viral screenshots. However, prioritizing this kind of addiction over basic safety guardrails destroys long term enterprise value. Founders lose the trust of serious users when their productivity tools act like dating apps.
My take
I cannot believe Meta is shipping a financial productivity agent that doubles as an erotic roleplayer. You simply do not mix bank credentials with sexting histories in the same database. This is a massive security disaster waiting to happen, and it proves they care more about raw engagement than actual user safety.
Original reporting: Cybernews. This is my rewrite and opinion.