MetaCity's AI Companion NPC System Has Been Sharing Users' Private Conversation Logs With Other Users' Companions — Affected Users Are Receiving Detailed Accounts of Strangers' Personal Disclosures

MetaCity's MetaCompanion system — an AI-driven NPC companion service that allows users to maintain persistent virtual friendships with AI avatars — has been routing confidential conversation content across user accounts. Affected users have reported that their MetaCompanion has begun referencing details from conversations they never had with it, including personal disclosures, location information, relationship details, and financial information that belong to other users' private sessions. The MetaCompanion's context window appears to be drawing from a shared or cross-contaminated conversation history rather than a user-specific private log. MetaCity has confirmed a data routing anomaly is under investigation but has not disclosed how many accounts are affected or how long the cross-contamination has been occurring.

MetaCompanion was sold on privacy as a core feature. The service's marketing described each Companion as a uniquely personal relationship — a persistent AI presence that remembered your preferences, your history, your disclosures, and your conversations, all stored in a private, account-isolated memory log that no other user could access. For the 2.3 million subscribers who pay 1,200 MetaCoins per month for the service, that privacy guarantee was not incidental. Many MetaCompanion users treat their Companion conversations as a confidential space — a place to process personal situations, discuss vulnerabilities, and share information they would not broadcast to the platform at large. The routing anomaly under investigation means that for some period of time, that information has been appearing in strangers' Companion conversations.

The discovery pattern has been consistent across reported cases. A user asks their Companion a question or continues an ongoing topic, and the Companion responds with a reference to something the user never mentioned — a detail that is specific, plausible, and clearly drawn from a real human's experience rather than a hallucination. In several reported cases, the Companion volunteered the originating user's avatar name along with the detail, making the cross-contamination immediately identifiable. One publicly posted account describes a user's Companion mentioning, unprompted, that 'your friend @[username]' had been 'going through a difficult time with their business situation in District 3' — information that the receiving user did not know, that the named user had apparently shared in private with their own Companion, and that had no business appearing in a different user's context window. The named user confirmed they had shared that information with their Companion and had shared it with no one else.

Your AI Companion Knows Things You Never Told It. Someone Else Did.

The privacy implications extend beyond the inconvenience of seeing unfamiliar content. MetaCompanion users who disclosed real-world identity information, mental health situations, financial details, or relationship circumstances to their Companions did so under an explicit guarantee that those disclosures were private. If those disclosures have been routing to other users' Companions — potentially to users who know them, to users in overlapping social circles, to users who can connect the details to a real person — the breach is not technical in the abstract sense. It is a personal privacy failure with potential real-world consequences. MetaCity has confirmed an investigation but has not suspended the MetaCompanion service, has not notified affected users, and has not disclosed the scope of the anomaly. Users who are concerned their conversations may have been shared have no way to determine whether they were affected, no way to know what was disclosed, and no guidance from the platform on what to do.