AI Agents on Mac: Apple Tightens Consent

Allowing an AI agent to access its files will still be possible on Mac. However, Apple wants this decision to be accompanied by a clearer understanding of the exposed data.
Apple anticipates increased risks with agent autonomy
The new protections do not constitute a ban: users retain the ability to open their files to AI agents. Loss of control over these agents has already occurred; such behavior could lead to the disclosure of private data or its misuse.
Last month, Apple assessed that the risks associated with such extensive access would significantly increase as these software programs gain capabilities and autonomy. The company is committed to explaining these risks before any authorization, to enable informed choices regarding personal data. It is notably marketing the Mac Studio to run AI models and agents locally.
A window details the permissions granted on Mac
Apple introduced controls in its latest beta of macOS, as presented on its developer blog. For applications such as ChatGPT, Claude, or Gemini, the company announces a process that requires explicit gestures before granting full disk access.
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisissez votre rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Screenshots shared on X by @madeby_ldt show a window warning that the Mac will no longer be able to protect data from the concerned application. The user must review and approve the authorized operations: reading, modifying, deleting, or exporting all data, as well as accessing hidden or recently deleted photos. Consulting the history and using the browser with the user's capabilities are also among these permissions.
This feature is available in the version designated as beta 27.3 of macOS 2. The rollout schedule for all users has not been communicated.
Meta contests access without consent
In September, Jason Aten, a columnist for Inc., reported that Muse, Meta's application, could access his messages even though he believed he had denied this permission. Meta responded that he had necessarily given his consent, asserting that this function depended entirely on consent. This episode caught Apple's attention.
Apple criticizes certain developers for the dangerous use of permissions that grant access to the entire disk. According to the company, files, emails, messages, and browsing history can thus be exposed without users fully understanding the implications. It has committed to addressing this issue through new controls.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.