Apple tightening macOS Full Disk Access controls over AI agent risks

By Splitfeed EditorialLast updated Oct 2, 2026, 19:30 UTC

On October 2, 2026, Apple announced it will add new controls around the Full Disk Access permission in macOS, directly citing the risks posed by increasingly capable AI agents. Full Disk Access is a broad system permission that grants applications reach over files, messages, mail, and browsing history. Apple framed autonomous AI agents operating on-device as a distinct and growing threat category, though the company has not yet released specifics on how the new controls will work or when they will arrive.

The announcement matters because it marks a public acknowledgment from one of the world's largest platform holders that AI agents represent a qualitatively different kind of software risk. For years, Full Disk Access existed as a rarely granted permission; as Martin Shelton noted in a reaction posted under the account @mshelton.bsky.social, such far-reaching permissions over a system and its data were once uncommon, but companies are now asking for them routinely. The genuinely contested question is whether tighter permission boundaries can keep pace with agents that are becoming more autonomous, or whether platform-level controls will prove insufficient against determined developers.

The Anti-AI camp has been most vocal. Shelton argued that regular people are already making very risky decisions by granting this access. John Spurlock, posting as @johnspurlock.com, said the risks associated with this level of access will grow substantially as agents become more capable and autonomous. Tyler King at @tyleraking.com connected the move to recent warnings about Meta's Muse agentic bot and device security. A more skeptical voice came from @cyberworm.uk, who questioned whether a workaround would emerge through the accessibility tree, pointing to what they described as a history of companies like Meta intentionally circumventing app permissions. The Middle Ground camp, represented here by Jason Broccardo at @zoocoup.bsky.social, called addressing the issue critical, treating Apple's response as a proportionate and necessary step. The Pro-AI camp has published no reactions to this story; typically, that camp would be expected to argue that capable AI agents require broad system access to be genuinely useful, and that over-restriction will blunt their value to users.

The detail most worth watching is the implementation Apple eventually releases, specifically whether the new controls impose granular per-agent permission prompts, restrict certain data categories outright, or take some other form. That specificity will determine whether the critics who expect workarounds are proven right, or whether Apple's intervention meaningfully changes how AI agents operate on its platform.