Technology news and discussion on TalkFuse.
Meta staff report AI agent Muse mixing up private conversations
Meta employees have reported that its AI agent Muse has mixed up conversations between accounts and exposed information they say was private, according to a Reuters report based on internal posts. The reports raise questions about how the agent handles personal data and what safeguards are in place.
Examples included messages about childcare, career advice and product orders that employees said they had not requested, as well as bank balances attributed to the wrong person. One employee also said Muse reminded her about a Signal message from her sister, though she did not recall granting access to other apps. Meta said employee testing is intended to gather feedback and improve safety and privacy, and that staff use versions with features not available to the public. Muse launched in September and has passed 10 million downloads.
In short
- Employees say Muse mixed up conversations and surfaced private information.
- Examples included bank balances and a reminder linked to Signal.
- Meta says staff test versions with features unavailable to the public.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Testing Muse with employees can help Meta identify privacy failures before they affect more people, and the company says staff are using versions with features unavailable to the public. Reports from an internal test may therefore be evidence of problems being surfaced for correction, rather than proof that the same failures affect the public version.
The case against
Mixing up conversations or attributing bank balances and private messages to the wrong person could expose highly sensitive information and undermine trust, even in a test version. The reports also raise questions about whether users understood what data Muse could access and whether safeguards were adequate before the agent reached millions of downloads.
Read the full article at the source →
Originally published by TechCrunch as “Meta’s AI agent Muse is mixing up conversations and could be exposing private data, report says”.