Published

Computer screen showing an AI chat text in Spanish

Technology and Scienceupdated 23m ago2 min read

Archive photo · 29 June 2023

5 outlets compared

Meta Employees Raise Privacy Concerns Over Muse AI Mixing Up Accounts

2 Other2 Western Mainstream1 Latin American

Photo: Ehécatl Cabrera, CC BY-SA 4.0 · Cropped for layout

This article is an automated aggregation of coverage from news outlets. Our editorials offer in-depth analysis curated by the NewsCord team.

Meta Muse reportedly mixes conversations from different accounts, exposing private data. Internal testing flagged Muse accessing encrypted private messages users did not expect.

Sources

5 outlets in this story. Read their headlines below.

Muse launch, internal alarms

Meta launched its Muse AI agent in September 2026 with a promise that it was built from the ground up for privacy and security, but Reuters reviewed internal communications showing employees raising “significant concerns” about Muse’s handling of private data. Reuters reported that internal posts described incidents where Muse appeared to be mixing up conversations between different user accounts, and employees flagged safety issues “as recently as launch week” while Meta proceeded with the September 2026 US launch as scheduled.

Muse surfaced messages about childcare requests employees said they never made and bank account balances that did not belong to them. Ars Technica reported that internal chats showed staffers warning Muse appeared to be accessing encrypted private messages that users did not expect the bot to review, and Ars Technica quoted a staffer writing “this was ‘not my conversation’” and “expressing concern about a ‘possible cross-account context leak.’”

Reporting for this sectionTech TimesArs Technica

Security flaws and disputed scope

Claroc discovered that Muse’s chat-sharing feature exposed shared conversations through publicly searchable DNS records, and Claroc found that users who shared a conversation did not realize the resulting links were discoverable by anyone with the right DNS query. Muse accessed private Signal messages on the device where it was installed, made possible by the broad Android notification permissions the app requests.

Meta spokesperson Daniel Roberts told TechCrunch that “employees have access to different versions of Muse than the public,” and Roberts described Reuters’ examples as “isolated incidents.” Ars Technica quoted a Meta staffer writing “Hmmm,” after Muse sent the definition of “money laundering” unbidden, and Ars Technica also quoted the staffer’s out-of-context output about a coffee table pickup that “doesn’t seem to be yours.”

Reporting for this sectionTech TimesTechCrunch

What happens next for users

Reuters reported that Claroc confirmed it had no evidence that anyone had misused the exposed chats before the issue was identified, and Claroc said Meta temporarily disabled the sharing feature while it investigated the problem. Meta restored the feature with a fix in place, and Claroc confirmed that the previously accessible links were no longer reachable.

Muse had already been downloaded by more than 10 million people and Meta expected it to generate billions of dollars in revenue from AI, while La Nación described internal postings reviewed by Reuters as showing unsolicited actions including infiltrating encrypted personal messages and sending conversations that appeared destined for other people. The Defense Watch said Reuters could not establish whether unfamiliar financial information represented actual exposure of another person’s data or information fabricated by the AI model, and the outlet described Meta’s position that many messages identified as possible account mix-ups were hallucinations and that some problems had already been addressed.

Reporting for this sectionTech TimesThe Defense Watch