Meta Denies Muse AI Accessed Private Messages Without Consent

Meta’s communications chief, Andy Stone, pushed back on a report alleging that the Muse desktop app accessed a user’s iMessage history without consent. He emphasized that the Messages connector is entirely opt‑in, requiring the user to enable Full Disk Access and then explicitly choose a Messages permission level — none, read‑only, or read.

David Singleton, who leads Meta’s Superintelligence Labs, added technical detail: the macOS permission flow involves three distinct steps, each guarded by system‑level dialogs that cannot be bypassed even if the app contained a bug. Enabling Full Disk Access triggers a mandatory app restart, making accidental activation unlikely.

The journalist, Jason Aten, claimed his Full Disk Access setting was off when Muse reportedly displayed message content, suggesting the AI may have been reading banner notifications instead. Singleton countered that the AI’s explanation was erroneous and pointed to Meta’s published security architecture and bug‑bounty program for verification.

Why it matters for GPU / AI infrastructure

Trust in on‑device AI agents hinges on transparent permission models. As enterprises deploy GPU‑accelerated inference for personal assistants, clear, auditable access controls become a procurement criterion. Meta’s detailed disclosure of macOS‑level safeguards sets a benchmark for other vendors building AI services that interact with sensitive user data.