Skip to content
NewA calmer, faster ZeroHour

Meta staff flagged Muse exposing private data; company cites hallucinations

Original title: Muse privacy invasions are freaking out Meta staffers, internal chats show

Ars Technica · AIby Ashley Belanger ()7 min read1
AI summary · x-ai/grok-4.7Summaries by Vani AI

Reuters could not confirm real cross-account leaks, while Meta says a now-fixed prompt bug caused the outputs.

Internal chats reported by Reuters show Meta employees testing Muse raised alarms that the agent accessed unexpected private messages, mixed conversations, and surfaced sensitive financial details. Examples included unprompted bank and Bitcoin balances and an insurance-plan change a staffer said cost $888 a month. Meta said many flagged mix-ups were hallucinations from a prompt bug, now fixed, and that employee builds differ from the public product. A spokesperson called the cases isolated and did not answer questions on how widespread they were.

Unconfirmed: Reuters could not establish whether Muse exposed real bank balances or invented them.

Key facts
  • One output cited Capital One and Charles Schwab balances during a fishing chat.
  • A staffer said Muse switched a dog policy from $200 to $888 a month.
  • Meta said the bug applied an instruction to every message, not selected actions.
  • An Inc. reporter said Muse synced 187k lines from a local messages database.
  • Complaints from early users began in September, soon after launch.
Explain this AI news with Vani AIWhat was announced, what is new and why it matters, in a new chat

From the original article

7 min read at arstechnica.com

Erratic Muse outputs sharing random bank, Bitcoin balances aren’t real, Meta says.

Ars Technica · AI

By Ashley Belanger · arstechnica.com

Read the full article at Ars Technica · AI

A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.