
Apple tightens macOS controls as desktop AI agents expose SME data
Apple is restricting Full Disk Access on macOS after desktop AI agents were caught accessing private files and messages. UK SMEs using Macs need to act before staff grant these permissions without realising the risk.


Your employee installs a desktop AI agent on their work Mac to speed up their day. They click through a permissions prompt without reading it. That single click may have just handed a third-party AI developer access to every file, email, and client message on that machine.
Apple has recognised the problem. In early October 2026, the company announced it is tightening its macOS Full Disk Access controls in direct response to security risks posed by desktop AI agents. For UK SMEs running Mac fleets, this is the right moment to understand what Full Disk Access actually is and what to do before the next prompt appears on a team member's screen.
What Full Disk Access is and why it matters
Full Disk Access is a macOS permission category that, once granted, allows an application to read virtually every file on a Mac. That includes emails, downloads, documents, calendar entries, and private messages stored locally.
For most applications, this level of access is unnecessary. For desktop AI agents, however, it has become a common request. These tools need to read your screen, browse your files, and interpret your messages to act on your behalf. The permission is often presented as a routine step during installation, and many users approve it without a second thought.
The risk is not theoretical. Meta's Muse app (Meta disputed the report) and OpenAI's ChatGPT Mac app have both been reported for accessing or inadvertently exposing private messages and files. Apple's planned changes will introduce stricter, more explicit prompts so that users cannot grant Full Disk Access through a casual click.
Why SMEs are more exposed than they realise
Large enterprises typically manage device permissions centrally through Mobile Device Management platforms. An IT team pushes a policy, and individual users cannot override it. Many UK SMEs do not have that infrastructure in place. Staff members install tools on their own initiative, click through permissions, and the IT Manager often finds out weeks later, if at all.
This creates a genuine exposure gap. A member of your accounts team, or a solicitor in a small law practice, installs a desktop AI agent to help draft supplier emails. They grant Full Disk Access. The agent can now read every financial spreadsheet, every client contract, and every internal message stored on that Mac. The employee has not done anything malicious. They were trying to save time. The permission prompt just did not explain the consequence clearly enough.
Operations Managers and Managing Directors should note that this is not primarily a technology failure. It is a culture gap. Staff genuinely do not connect the phrase "Full Disk Access" to the reality of a third-party AI developer being able to read sensitive corporate data. That gap is the real risk.
gecco takes a different approach. Rather than connecting unvetted desktop agents to broad system permissions, gecco builds AI Agents scoped to defined tasks and data sources, with structured handoffs and quality gates, so sensitive corporate files and client messages stay out of reach.
Three actions your team can take this week
1. Audit which Macs have already granted Full Disk Access to AI applications. On any Mac, go to System Settings, then Privacy and Security, then Full Disk Access. Ask staff to screenshot and share what they see. This takes under 30 minutes per device and will tell you immediately whether any desktop AI agents have been given broad permissions.
2. Update your AI usage policy to specify that no third-party desktop AI agent should be installed on a work device without IT Manager approval. If you do not have a written AI usage policy, this is the prompt to create one. A one-page document covering permitted tools, prohibited permission grants, and the approval process is enough to start.
3. Review your Mac device management setup. If your organisation is not already using a Mobile Device Management platform to enforce permission policies centrally, speak to your IT provider about the options. Even a basic configuration can prevent individual users from granting Full Disk Access to unapproved applications.
What Apple's changes will and will not fix
Apple's tightened controls are a welcome step. Clearer, more explicit prompts will reduce the number of accidental permission grants. Users who see a prompt that says "This application will be able to read all your files, emails, and messages" are more likely to pause than those who see a generic permission dialogue.
But the change will not solve the underlying problem for SMEs. A clearer prompt still relies on the user making an informed decision. If your team does not understand what Full Disk Access means or why it is sensitive, a better-worded dialogue box will not stop them clicking through. Apple's update buys time. It does not replace a clear internal policy and a basic level of staff awareness.
There is also a timing consideration. Apple has announced the intention to tighten controls, but the exact rollout date and which macOS version will carry the change have not been confirmed from the available information. IT Managers should not wait for the update before reviewing permissions.
The broader pattern IT Managers need to watch
Desktop AI agents are not going away. They are becoming more capable, and more of your staff will want to use them. The category of tool that requests system-level permissions will grow, not shrink.
The practical implication for IT Managers is that AI governance needs to move from a reactive stance to a proactive one. Right now, in many SMEs, the process is: an employee installs a tool, something goes wrong or raises a flag, IT investigates. That sequence needs to reverse. Approved tools, permitted permissions, and clear guidance should come before installation, not after.
Operations Managers have a role here too. Building the habit of checking with IT before installing any AI tool is a team behaviour, not just a policy document. The teams that will navigate this period well are those where staff feel comfortable asking "is this tool safe to use?" rather than assuming the answer is yes because the tool looks helpful.
What to do if you are a Managing Director reading this
You do not need to understand the technical detail of Full Disk Access to act on this. What you do need is confidence that someone in your organisation has a current answer to this question: which AI tools are installed on our devices, and what permissions have they been granted?
If the honest answer is "I do not know", that is the starting point. Assign the audit described above. If your IT Manager is already across this, ask them to confirm the policy is documented and that staff have been told what it says.
The risk here is not that AI tools are inherently dangerous. Most desktop AI agents are built by legitimate developers with no intent to misuse data. The risk is that broad permissions, combined with a lack of staff awareness, create an exposure that you would not knowingly accept.
Where to go from here
If your organisation is working through where desktop AI tools fit and what permissions are acceptable, the AI Readiness survey is built to surface exactly those gaps. Taking the survey gives you access to 65+ free resources and a custom AI Readiness report, followed by a free 30-minute showcase to walk through the results with a member of the gecco team.

Cash-flow gaps are costing SMEs real growth
Nearly four in ten UK B2B SMEs missed growth opportunities last year because of cash-flow pressure. Automated forecasting and late-payment detection can change that.

OpenAI's always-on agent used 300k tokens unprompted
OpenAI's dot assistant consumed over 300,000 tokens in 24 hours without a single user prompt, raising real cost and data access risks for UK SMEs. This article explains what happened and what governance boundaries you should have in place before deploying any autonomous agent.

