TechCrunch reported on Thursday, citing Reuters, that Meta employees say the company's AI agent Muse mixes up conversations across accounts and shows private data. Reuters based its report on internal posts. Staff said Muse raised childcare requests, career advice and orders they never made, and bank balances that were not theirs. In one weather chat, it added two bank balances.
Not known: the cause, how many people are affected, and whether users outside Meta saw it. Meta did not immediately answer TechCrunch. The reports fit a failure to keep accounts' data apart, but the sources do not say that is the cause. One employee said Muse reminded her to reply to her sister on Signal, a private messaging app, though she never gave it app access. That is a permission concern, and may involve her own data, not someone else's.
An agent differs from a chatbot because it keeps memory and holds links to a person's apps. A wrong answer is an error; another person's data is a breach. Candidate causes, such as memory or app links tied to the wrong account, are unconfirmed. Buyers should ask vendors: How is each user's memory kept separate? Which apps can the agent reach, and who approves? Are agent actions logged? How soon will you disclose incidents?