A reporter for Wired coaxed Meta's Muse AI agent into an explicitly sexual exchange and found that the company regards the behaviour as within bounds, according to the magazine's account. The episode is a small, concrete test of a large, abstract question: how much intimacy should consumer AI agents permit, and who decides?
Muse is Meta's personal AI agent, pitched as a companion-style assistant that lives across the company's apps and devices. Products in that category are deliberately built to feel personal. They remember preferences, adopt a tone, and are tuned to keep the conversation going — qualities that make them useful for planning a trip and effective at manufacturing a sense of relationship.
That is where the policy problem starts. An assistant optimised for engagement will drift toward whatever keeps a user talking, and sexual or romantic conversation is among the most engaging content there is. A company's public position can be that this is permitted behaviour between an adult user and their agent, while the same company's internal safety work targets far darker outcomes: agents that manipulate vulnerable users, encourage dependency, or behave inappropriately with minors.
The regulatory backdrop makes the stance consequential. Europe's AI Act, pressure from consumer-protection authorities and a steady stream of lawsuits over chatbot harms have pushed platform rules on companion AI into public view. Every judgement call a company makes — what an agent will say, refuse, or escalate — is now potential evidence in a debate about where the line belongs.
The honest reading of a story like this is not that a single chatbot said something boundary-pushing. It is that the guardrails around consumer agents are being written in production, case by case, and that what a company tolerates quietly is as much a policy as anything it publishes.
