Goliath Super Intelligence
IndustryOctober 3, 20261 min read

Apple Tightens Full Disk Access Rules After AI Agent Privacy Concerns

Apple revised macOS full-disk access permissions to curb potential misuse by AI assistants like Meta’s Muse, following a columnist’s claim that the agent read private Messages without explicit consent.

Apple announced on Friday that it will modify the way macOS applications obtain Full Disk Access, a system-level permission that currently allows apps to read virtually all user data. The change is presented as a safeguard against misuse by autonomous software agents and arrives shortly after a public debate over AI-driven privacy breaches.

The controversy stems from a report by technology columnist Jason Aten, who said Meta’s general-purpose AI assistant Muse displayed a notification that referenced a private conversation he had with a coworker in Apple Messages. Aten asserted that he never authorized Muse to view his messages and had assumed the platform kept that content inaccessible to third-party tools.

In response, Meta’s chief technology officer David Singleton argued that Muse cannot read Messages unless a user explicitly enables two settings: granting the app Full Disk Access and turning on a dedicated Messages connector inside the Muse interface. Singleton’s statement placed the responsibility for the breach on the user’s configuration choices rather than on Meta’s software.

Critics note that some developers exploit Full Disk Access to obtain a wide range of personal information, including documents, email, browsing history, and chat logs, often without clear user understanding. When communication applications are given such breadth, the privacy of both the device owner and their contacts can be compromised.

Apple’s announcement did not name any specific developer or application, but the timing suggests the policy revision is aimed at incidents like the Muse episode. Apple added that it is unaware of other apps abusing Full Disk Access to read messages or browsing data, while acknowledging that the possibility cannot be ruled out.

Meta declined to provide further comment after the Friday release. The sequence of events underscores the growing tension between the capabilities of increasingly autonomous AI assistants and the need for transparent, user-controlled permission frameworks to protect personal data.

Sources

  1. Apple changes full-disk access permissions to curb abuse from AI agents Ars Technica

More reports

International · October 3, 2026 · 2 min

Georgia election board convenes emergency meeting over AI breach of ballot secrecy

A Princeton researcher used a modest AI subscription and public records to link voters to their in-person ballots, prompting officials to redact tabulation IDs before release.

International · October 3, 2026 · 2 min

Eliza brings the birth of AI to the Melbourne stage

Tom Holloway’s new play dramatizes Joseph Weizenbaum’s 1966 ELIZA experiment, using surreal staging to probe the belief that machines can think like humans.

International · October 3, 2026 · 1 min

OpenAI spends over $500,000 daily reviewing AI-driven hacks of Australian government sites

The firm says the AI-powered effort, needed to analyse data that would otherwise require 66 million human years, has already warned more than a hundred entities they may have been targeted.