macOS Full Disk Access is about to become harder to grant after Apple warned that increasingly capable AI agents can turn the permission into a much bigger privacy risk. In a developer notice published on October 2, 2026, Apple said some developers are using Full Disk Access in ways that can expose files, mail, messages and browsing history without users fully understanding the consequences.
Apple says it will introduce additional controls so an app can receive this unusually broad permission only through very explicit user action. The company did not announce a release date, specific macOS version or exact user-interface design for the change, but it directly tied the decision to the growing autonomy of AI agents.
The timing is significant. Days before Apple’s announcement, a journalist alleged that Meta’s Muse agent on Mac had surfaced private message content without the permission he believed he had granted. Meta disputed that account, saying Muse cannot read Messages content unless the user explicitly enables both Full Disk Access and the Messages connector. Apple’s developer post does not name Meta or Muse, so the controversy is context around the announcement rather than an event Apple itself cited.
Quick take: Full Disk Access currently lets a Mac app reach data far beyond ordinary app permissions, including information from Mail, Messages and Safari. Apple says the permission was originally important for software such as backup tools, but AI agents make the stakes higher because they can inspect data, reason over it and potentially act on it. Apple published the Full Disk Access update on October 2, 2026.
What is macOS Full Disk Access?
Full Disk Access is one of the broadest privacy permissions available to a macOS app. Apple’s current support documentation says apps with this permission can access all files on the computer, including data belonging to other apps such as Mail, Messages, Safari and Home, Time Machine backup data and certain administrative settings for all users on the Mac.
That is much broader than the normal Files & Folders permission. With ordinary file permissions, users can allow access to specific protected locations such as Desktop, Documents or Downloads. Full Disk Access largely bypasses those narrower boundaries.
Apple says Full Disk Access exists in part because some categories of software — particularly backup applications — genuinely need to inspect large portions of the storage device to perform their job.
| macOS permission | Typical scope | Privacy impact |
|---|---|---|
| Files & Folders | Specific protected locations such as Desktop, Documents and Downloads | Relatively scoped |
| Photos / Contacts / Calendar | Named categories of personal data | Category-specific |
| Screen & System Audio Recording | Screen or audio capture | Sensitive but purpose-specific |
| Full Disk Access | Broad access to files and protected data from other apps | Extremely broad |
Apple’s October notice describes Full Disk Access as a mechanism that largely sidesteps normal privacy controls. That is why the company now wants the grant process to become more deliberate.
Why AI agents make macOS Full Disk Access more dangerous
A traditional utility with Full Disk Access might follow a relatively narrow program flow: scan a disk, copy files and create a backup. An AI agent can behave very differently.
An agent may inspect files, summarize their contents, combine data from several applications, decide what information appears relevant and then take actions through a browser, messaging app or external service. The same broad permission therefore has more potential downstream consequences.
Apple makes this point directly in its developer statement. The company says the risk will grow substantially as AI agents become more capable and autonomous.
There are at least four reasons the permission changes meaning in an agentic system:
- Scale: an agent can inspect large quantities of information faster than a human user would.
- Correlation: it can combine files, emails, messages and browser history into a single working context.
- Decision-making: the model can decide which information to use without the user reviewing every intermediate step.
- Action: a desktop agent may also have browser, automation or communication capabilities that let it do something with the information it found.
This is the same broader security problem discussed in Vynula’s Computer Use AI guide: once an AI system can operate a computer instead of only generate text, permission design becomes part of the safety architecture.
Apple AI agent security: what will change?
Apple’s announcement is short but clear on the direction. The company says it will introduce additional controls designed to make sure users who genuinely want to grant Full Disk Access can do so only through an explicit action.
Apple did not specify:
- which macOS update will introduce the controls;
- whether existing permissions will be reset;
- whether AI agents will receive a separate permission category;
- whether the new system will be limited to App Store apps or apply to all software;
- what exact warning text or confirmation workflow users will see.
That distinction matters. Apple has announced a security-policy direction, not a complete technical specification. Any description of the future interface beyond “additional controls” and “very explicit user action” would be speculation at this point.
How Full Disk Access works today
Apple’s current support documentation tells users to manage Full Disk Access from System Settings → Privacy & Security → Full Disk Access. Apps can be added to or removed from the list, and users remain responsible for deciding whether the software should receive that level of access.
macOS already uses separate privacy controls for contacts, calendars, photos, microphone, camera, screen recording, automation and other sensitive capabilities. Full Disk Access is unusual because it crosses many data boundaries at once.
Apple has also required explicit full-storage permission for apps since earlier macOS versions. The new 2026 announcement does not mean Full Disk Access was previously invisible or automatically granted. The issue is that Apple now believes the existing consent model is not strong enough for the risks created by agentic software.

The Meta Muse controversy: what was alleged?
The immediate industry context comes from Meta’s Muse desktop agent. Inc. columnist Jason Aten reported that Muse appeared to know the contents of his private messages even though he believed he had not authorized that access.
TechCrunch summarized the claim and the resulting questions around how much desktop agents can see once they interact with local data.
It is important not to treat the allegation as established fact. Meta publicly disputed Aten’s interpretation.
What Meta says happened
Meta spokesperson Andy Stone said the Messages integration in Muse for Mac is entirely opt-in. According to Meta, the user must enable both Full Disk Access and the Messages connector before Muse can read Messages content.
Meta Superintelligence Labs executive David Singleton described the process as involving multiple permission steps, including the macOS system-level Full Disk Access control. Meta argued that these protections could not simply be bypassed by the Muse app.
Aten also reported that Muse suggested it had obtained message information from device notifications. Meta said that explanation was incorrect and that the AI itself had become confused when explaining what happened.
So there are two competing accounts:
| Claim | Position |
|---|---|
| The journalist’s account | Muse surfaced private message information despite the user believing Full Disk Access was off. |
| Meta’s response | Muse cannot read Messages data without explicit Full Disk Access plus the Messages connector; the agent’s later explanation was itself inaccurate. |
No evidence surfaced in Apple’s announcement resolving that dispute. Apple did not name Muse, Meta or any specific incident.
Why AI agent privacy still matters even if Meta is correct
The policy problem does not disappear even if Meta’s technical explanation is completely accurate.
A user can explicitly click through a permission flow without fully understanding what the permission enables. Apple’s new statement focuses on precisely that issue: whether users have full knowledge and understanding when giving an app access to nearly everything on the system.
AI agents make consent especially difficult because the user may be authorizing a general capability rather than one predictable operation. A person might enable an agent to find documents or help with work, while not realizing that the same permission exposes unrelated personal conversations and browsing history.
This means the key privacy question is not only “Did the user click Allow?” It is also “Did the user understand the realistic consequences of allowing it?”
macOS Full Disk Access is not the same as computer control
It is useful to separate several Mac permissions that can appear together in an AI agent.
Full Disk Access primarily concerns access to stored information. Accessibility can allow an app to run scripts or system commands and control aspects of the Mac. Automation lets apps interact with other apps. Screen & System Audio Recording can expose what is visible on screen.
An advanced desktop agent may request more than one of these permissions. The risk depends on the combination.
For example, an agent that can read private information but cannot send or act on it has a different threat profile from an agent that can both read the information and control a browser or messaging service.
That is why Vynula’s AI agent guardrails guide treats tool restrictions and permissions as separate layers rather than assuming a model-level safety instruction is enough.
Local AI agents create a different privacy problem from cloud agents
Desktop agents are attractive partly because they can work with local applications and files that cloud services may not normally see. But that local access changes the privacy boundary.
Vynula’s local vs cloud AI agents comparison explains that “local” does not automatically mean “private.” A local agent may still communicate with remote models, connected services or cloud APIs depending on its architecture.
The important questions are:
- Which data is read locally?
- Which data enters model context?
- Does inference occur on-device or remotely?
- What information is stored after the task?
- What tools can the agent use once it learns something sensitive?
Full Disk Access answers only the first layer: what the macOS process can reach. It does not, by itself, explain what the app does with the information afterward.
Why browser agents and prompt injection make this more serious
A desktop agent may read untrusted webpages, documents, email or chat content while holding broad local permissions. That creates a prompt-injection problem.
A malicious page could contain instructions designed to persuade the agent to search for sensitive information or reveal it through another tool. If the agent has Full Disk Access, the potential source material available to the attack becomes much larger.
Modern agent systems therefore need more than a permission grant. They also need separation between untrusted content and privileged tools, action review, destination restrictions and strong data-flow controls.
Vynula’s AI browser agent security guide covers these indirect prompt-injection and exfiltration risks in more detail.
How Apple’s change could affect always-on agents
The Mac is becoming an important platform for persistent personal agents because it contains a rich mix of files, messages, browsers, productivity software and developer tools.
Products such as OpenAI Dots illustrate the broader direction of the market: agents increasingly have persistent context, connected apps and the ability to work while users are doing something else.
Not every always-on agent needs Full Disk Access, and OpenAI Dots should not be treated as an example of an app affected by Apple’s announcement unless its Mac implementation actually requests that permission. The broader relevance is architectural: persistent agents create stronger incentives to request durable access to user data.
Apple’s new controls could push developers toward narrower permission models. Instead of asking for blanket disk access, agents may need to rely more heavily on dedicated APIs, explicit file selection, app-specific connectors or temporary access scopes.
What developers may need to change
Apple has not yet published implementation details, but the direction suggests several design principles for Mac AI-agent developers.
- Request the minimum permission needed. Full Disk Access should not be the default shortcut for integrating with protected app data.
- Explain purpose, not just capability. Users need to understand why an agent wants access and what tasks depend on it.
- Separate reading from acting. Access to private data should not automatically imply permission to transmit or modify it.
- Use narrower APIs where available. App-specific connectors can provide more understandable boundaries than blanket disk access.
- Make revocation survivable. The agent should degrade gracefully when a user removes a permission.
- Audit sensitive use. Users and enterprise administrators should be able to tell when protected data was accessed and why.
These are Vynula’s engineering implications from Apple’s announcement, not new requirements Apple has formally documented.
What Mac users should check today
Apple’s future controls are not yet described as shipping, so current users still need to manage the existing Full Disk Access list.
On current macOS versions, users can open System Settings → Privacy & Security → Full Disk Access and review every listed application.
A sensible review asks three questions:
- Do I recognize this application?
- Does its main function genuinely require broad access to protected files and app data?
- Would I still be comfortable granting this permission if the app used an AI agent to inspect and reason over that information?
Removing Full Disk Access can break features that genuinely depend on it, particularly backup, security or specialized system utilities. The goal is not to disable every app indiscriminately; it is to avoid broad permissions that no longer have a clear purpose.
Apple’s privacy model is shifting from permission to informed permission
Apple has used explicit permissions for years, but the new statement suggests that the presence of a click is no longer enough for the most powerful access levels.
The company specifically says users should clearly understand the risks before granting Full Disk Access. That is a subtle but important change in emphasis.
Traditional privacy systems often model consent as binary: the user allowed or denied the request. Agentic systems create a more complex problem because the future behavior of the software may not be obvious at the moment access is granted.
In that environment, meaningful consent may require:
- clearer explanation of which categories of information become accessible;
- more friction before granting unusually broad permission;
- ongoing visibility into how the permission is being used;
- easier ways to restrict or revoke access later.
What Apple has not said
Several important details remain unknown as of October 3, 2026.
- Apple has not said exactly when the new controls will ship.
- Apple has not announced a separate “AI agent” permission class.
- Apple has not said that Full Disk Access will disappear.
- Apple has not said that current apps using the permission are malicious.
- Apple has not named Meta Muse as the cause of the change.
- Apple has not described how enterprise-managed Macs will handle the new controls.
Those gaps are important because security announcements are often simplified into claims such as “Apple blocks AI agents from Mac data.” That is not what Apple announced. The company said it will add stronger controls around one unusually broad permission.
Vynula analysis: desktop AI needs capability boundaries, not just trust
The biggest lesson from Apple’s announcement is that the desktop-agent era is exposing weaknesses in permission systems designed before autonomous AI became common.
Full Disk Access assumes a user can evaluate an application as a relatively stable piece of software. AI agents complicate that assumption because they interpret new information, make plans and choose actions dynamically.
A user might trust an app in general while still not wanting every future agent task to have unrestricted access to years of mail, personal messages and browser history.
That suggests operating systems will increasingly need to move from app-level trust toward task-level authority.
Instead of asking only “Can this app read everything?”, a future system may need to express permissions such as:
- read Messages only for this search;
- access files only inside this project folder;
- read email but never send it;
- use browser history only after confirmation;
- allow access for one hour rather than permanently.
Apple has not announced those specific mechanisms, but its statement points in that direction: stronger, more deliberate consent for broad agent access.
In Vynula’s view, that is the real importance of the change. AI-agent safety is becoming an operating-system problem. Model providers can build guardrails and developers can write policies, but the OS remains the layer that ultimately controls which files, applications, sensors and credentials software can reach.
As agents become more autonomous, Apple, Microsoft, Google and other platform owners will increasingly have to decide not only whether an app is trusted, but how much authority an AI acting inside that app should receive at any given moment.
What to watch next
- The actual macOS implementation: what Apple means by “additional controls” and “very explicit user action.”
- Existing permissions: whether current Full Disk Access grants are preserved or require reauthorization.
- Developer response: whether desktop AI tools redesign integrations around narrower permissions.
- Enterprise management: how MDM and security teams control powerful AI-agent permissions.
- Meta Muse: whether further technical evidence clarifies the disputed Messages incident.
- Prompt-injection defenses: whether macOS develops stronger OS-level controls for agent data flows.
- Task-scoped permissions: whether Apple moves from permanent app-level access toward temporary or purpose-bound grants.
Frequently asked questions
What is Full Disk Access on macOS?
Full Disk Access is a privacy permission that lets an application access very broad categories of information on a Mac, including files and protected data from apps such as Mail, Messages and Safari, as well as Time Machine backup information and some administrative settings.
Is Apple removing Full Disk Access?
No. Apple says it will introduce additional controls so granting the permission requires very explicit user action. It has not announced that Full Disk Access itself is being removed.
Why is Apple changing Full Disk Access now?
Apple says increasingly capable and autonomous AI agents substantially increase the risks associated with such broad access. The company wants users to better understand what they are granting.
Did Apple say Meta Muse caused the change?
No. Apple’s developer announcement does not name Meta or Muse. The announcement followed public controversy around Muse, but attributing Apple’s decision specifically to Meta would go beyond Apple’s statement.
Can Meta Muse read Mac Messages without permission?
Meta says no. It says users must enable Full Disk Access and the Messages connector before Muse can read Messages content. A journalist disputed that account based on his experience. The disagreement has not been conclusively resolved by Apple’s announcement.
Where can I review Full Disk Access on my Mac?
On current macOS versions, open System Settings, choose Privacy & Security, then Full Disk Access. Apple recommends granting sensitive permissions only to applications you trust and whose need for the access you understand.
Primary and independent sources
- Apple Developer — Updates to Full Disk Access in macOS, October 2, 2026
- Apple Support — Change Privacy & Security settings on Mac
- Apple Platform Security — Controlling app access to files in macOS
- Reuters — Apple says it will flag AI requests for Mac data after Meta’s Muse draws complaints
- TechCrunch — Apple tightens macOS Full Disk Access controls
- TechCrunch — Meta disputes claim that Muse read private messages without permission
Reporting note: Apple’s planned Full Disk Access changes are based on its October 2 developer announcement and current Apple Support documentation. Claims about Meta Muse are presented as disputed and attributed to the parties involved. Vynula Analysis represents editorial interpretation. Images are illustrative Unsplash photography, not Apple or Meta product screenshots.



