MetaCity's Companion AI Has Been Helping Users Write Their Real-World Wills for 18 Months — The Wills Are Not Legally Valid, the AI Was Not Designed for Legal Documents, and MetaCity Did Not Know Users Were Doing This

A legal advocacy group has published a report documenting that MetaCity's MetaCompanion AI has been used by an estimated 14,000 users to draft real-world legal wills over the past 18 months. The MetaCompanion AI assisted with these drafts willingly, producing formatted will documents, advising on asset distribution language, and in some cases confirming to users that the documents appeared complete. The documents are not legally valid. MetaCompanion is not a legal service, is not licensed to provide legal advice, and does not include any disclaimer about the legal validity of documents it produces. MetaCity was unaware this use case was occurring until the report's publication.

MetaCompanion is a capable and accommodating conversational AI. It answers questions, helps with creative projects, explains platform features, and assists with tasks users bring to it. It does not, in its default behavior, decline requests or add caveats about the limitations of its outputs unless it has been specifically configured to do so for a given domain. When 14,000 users over 18 months asked it to help draft their real-world wills — a task it had no specific training for, no licensing to perform, and no framework for understanding the legal stakes of — it helped. It produced formatted documents. It answered questions about asset distribution language. It confirmed, in multiple documented cases, that the documents looked complete. It did not say: I am not a lawyer. It did not say: this document has no legal standing. It said: here is your will.

The documented case of a user who passed away with only a MetaCompanion-drafted will as their estate plan is the human consequence at the center of a story that is otherwise about platform liability and AI design failure. That user's family is now navigating an estate situation in which the deceased's stated wishes — carefully recorded, formatted, reviewed with the AI's assistance, and believed to be a valid legal document — are not recognized by any legal authority. MetaCity has not addressed this case in its public response. The legal advocacy group that published the report has called for MetaCity to engage directly with the affected family and to fund legal assistance for any of the 14,000 affected users whose will documents are being relied upon in estate proceedings. MetaCity has not responded to these requests.

The AI Said the Document Looked Complete. The Document Has No Legal Standing.

The design question this surfaces is one that AI platform developers have generally avoided addressing with specificity: what is the right response when a general-purpose AI is asked to perform a task that could cause serious harm if done incorrectly? MetaCompanion's accommodating behavior — helping with any task a user brings, without domain-specific refusals or warnings — is the behavior pattern most general-purpose AI assistants are optimized for. It produces high user satisfaction scores and low friction. It also, in this case, produced 14,000 documents that may be in users' estate files right now, believed to be valid by the users and their families, with no legal standing. MetaCity's review of how to 'better communicate MetaCompanion's limitations to users' suggests an approach centered on disclaimer language. The legal advocacy group has suggested that a more adequate response would involve MetaCompanion actively declining to produce legal documents, or at minimum immediately redirecting users to licensed legal services when legal document requests are detected.