AIAIAI.FYI
LIVE

Meta employees say Muse AI agent sent unprompted messages citing private data, Reuters reports

Meta says the odd messages were hallucinations and that Android notification access is opt-in. Muse has passed 10 million downloads since its September launch.

What happened

Meta's AI agent Muse sent some Meta employees messages they did not ask for. One staffer got a reminder to reply to her sister on Signal, an encrypted chat app. Another got bank balances in a chat about fishing. Meta says those balances were hallucinations, meaning the AI made things up.

Why it matters to you

AI agents that read your phone or apps can seem to know private things. People may not know what access they have granted.

The takeaway

Check which permissions you give an AI agent, especially notification access.

What it means for you

If you build software

Engagement prompts that run between sessions can make an agent invent sensitive details, and default notification access widens what it can see.

If you run a business

JPMorgan Chase will not let agents access customer accounts, and Amazon blocks Muse from shopping on its site, so agent access is contested.

If you invest

Fortune reports Muse is key to Meta's plan to earn billions of dollars from AI, and the privacy complaints come from its own staff.

Get stories like this every morning