Meta's Muse AI References Employee's Signal Message

Meta says incidents in which its AI assistant Muse produced unrequested or inaccurate messages are largely hallucinations, and internal concerns about its permissions were raised.

Updated on Nov. 19, 2026 in Artificial Intelligence

Meta's Muse AI References Employee's Signal Message

Live Poll

Would you trust an AI assistant with access to your private messages?

Meta's AI assistant Muse referenced a Signal message a Meta employee had received from a family member, though the employee had not intentionally connected Signal to Muse. Meta says Muse can access only data on devices users have given it permission to access.

Why it matters

Muse is designed to send emails, make calls, and run searches on users' behalf, so questions about its device access and permissions bear directly on how much control users retain over their personal communications. Meta says its Terms of Service hold users responsible for actions their agents take.

Muse agents request deep access to users' devices, and Meta says the assistant is limited to data on devices where users have granted permission. The extraction does not specify the technical mechanism by which the Signal message was accessed.

The players

Meta

Meta is the technology company that owns Facebook, Instagram, and WhatsApp and is developing the Muse AI assistant.

The details

Meta employees reported receiving Muse messages that appeared to answer queries they never sent, and internal posts about these incidents were discussed from mid-September through November. Meta described one incident in a late September social media post as a one-day bug fixed the same day, while stating that many such incidents are hallucinations.

Timeline

  1. Late September 2026: a Meta executive described one Muse incident as a one-day bug in a social media post.

  2. Mid-September through November 2026: Meta employees discussed Muse message incidents internally.

The Tech Race

The Muse incidents illustrate the risks that accompany agentic AI assistants designed to act across a user's devices, a design direction pursued by several major technology companies competing to build such tools. Meta's position that users bear responsibility for their agents' actions places the burden of oversight on individuals as this category of product expands.

Users who grant AI assistants access to their devices may have messages, searches, and other data reached in ways they did not intend. Meta's Terms of Service state that users are responsible for actions their agents take, so reviewing the permissions granted to such tools is a practical step.

The takeaway

Granting an AI agent broad device permissions can expose private messages to the assistant, so reviewing and limiting those permissions is a practical safeguard. Users should also recognize that under Meta's terms they may be held responsible for what their agents do.

Further reading

Follow the latest developments in Artificial Intelligence coverage for context on AI assistant design and risks.

Source note: This article includes information reported by Reuters.

Live Poll

Would you trust an AI assistant with access to your private messages?