EN ▾
SNAPE REP RT Front page →
DevelopingAI safety21 Nov, 02:39 UTC

Meta's Muse AI agent has sent users others' chats and read Signal messages, Reuters finds

A commuter walks past Meta Muse ads displayed on screens at a subway station in New York City.
REUTERS/Jeenah Moon, via Astro Awani

Reuters reviewed more than two dozen internal posts and, with Signal and the AI Now Institute, replicated access to Signal messages; Meta calls the examples "isolated incidents".

Read at Astro Awani International (Reuters) Share

Why it matters

Muse is described as one of the fastest-growing apps on record, with more than 10 million downloads since its September launch, and is central to Meta's plan to earn billions from AI. The internal posts describe the agent mixing up accounts: employees say it sent them bank balances, childcare requests and orders that were not theirs, and that it had default access to Android push notifications. Meta says the cases are isolated and in part hallucinations. The finding bears on how much access consumer agents should be given by default. Watch for a Meta fix, regulator interest and whether others can reproduce the Signal result.

Key facts

  • Reuters reviewed more than two dozen internal Meta posts
  • Muse has more than 10 million downloads since its September launch
  • Staff reported others' bank balances and orders sent to them
  • Muse had default access to Android push notifications, per one employee
  • Meta spokesperson Daniel Roberts: "isolated incidents"

Sources

More on AI safety
More from Snape ReportFront page →