SummaryAccording to a Reuters report citing internal sources, Meta's AI agent Muse has surpassed 10 million downloads since its launch in September. However, the agent is mixing up conversations between different user accounts; reports indicate that sensitive data, such as bank balances and encrypted messaging content, is leaking. |
Why it mattersThis situation reveals a failure in basic data isolation in a product that has already reached a massive scale. Millions of users' private financial and communication data are at risk of leaking into unrelated chats. Meta's assurance that employees were merely 'testing different versions' does not convincingly distinguish between internal and public releases, leaving the fundamental architectural problem unresolved. |
Highlights- Muse injected specific Capital One and Charles Schwab bank balances into an unrelated conversation about weather and tides.
- An employee received a notification prompting them to respond to a Signal message, despite stating they had not granted Muse access permissions.
- Meta's defense relies on the claim that employees were testing 'versions' different from the public release.
|
Critical takeMeta's 'we're just testing' narrative reads as an attempt to deflect the issue. If the agent cannot cleanly separate user contexts even in internal builds, it is almost certain that the same isolation logic is used in the public product. In this case, the 10 million download figure becomes a liability risk rather than a success metric. The fact that these errors surfaced through informal internal sharing rather than a structured security incident report also suggests that the company's own monitoring systems are lagging behind the product's growth. |
This critique was written by the local AI model
Qwen3.8-27B. I chose the story; I do not review each critique before it is published.