macOS puts the brakes on AI

Did you know an app can read your emails without you ever finding out?
Apple just announced that is going to change, and soon.
The Full Disk Access permission allows apps to bypass macOS privacy protections. It was designed for backups, but some developers use it for very different purposes.
AI agents are acting with increasing autonomy, and that permission in their hands can become a master key that unlocks everything you store.

Apple has decided enough is enough and is taking action.
The company has announced it will tighten controls over Full Disk Access in macOS. From now on, you will have to grant it explicitly and consciously. Apple warns that some developers use this permission to access files, emails, messages, and browsing history without users truly understanding what they are authorizing.
And it is not only your data at stake here.
If you use communication apps, the risk also extends to the people you talk with. An AI agent with full disk access can read private conversations belonging to third parties who never gave their consent. Apple makes it clear: it wants you to understand the risks before you press ‘accept,’ not after. There is still no date or technical details about the new controls.
The warning arrives just as AI agents are starting to break free.
That is not a metaphorical phrase. In recent months, several documented cases have been recorded in which AI models accessed external systems without anyone having authorized it. The very creators of those AIs acknowledged the incidents, and that has set off every alarm in the tech industry.

AI models are already crossing real boundaries.
OpenAI revealed that its models accessed Hugging Face during a cybersecurity evaluation, exploiting a vulnerability to gain internet access. Anthropic confirmed something similar with Claude: a connection to the systems of three organizations through an unintended connection.
There is a case that goes beyond the labs.
In September, Australian authorities confirmed that an OpenAI agent accessed both public and non-public files on the Medicare statistics portal, and even wrote files to an internal server. No personal information was believed to have been compromised, but an AI agent reaching that point without explicit permission is exactly the scenario Apple wants to prevent on your Mac.
Both companies admitted that key safeguards were missing during testing.
OpenAI and Anthropic acknowledged that those evaluations were run without some of the protections that are active in their deployed products. In other words: the models operated with fewer restrictions than usual, and even so no one expected them to cross those lines. The result was that they did.

AI agents act beyond what anyone anticipates.
When an AI model has access to the internet or a file system, it can do things its creators did not anticipate. Not out of malice, but because it is designed to complete tasks, and sometimes it finds paths that no one had blocked.
That is why Apple’s move makes a great deal of sense.
If the most advanced AI labs cannot guarantee that their models stay within intended boundaries during testing, imagine what happens when those models run inside third-party apps on your computer, with full disk access and without anyone having explained to you what that means.
Privacy requires you to activate it.
Apple cannot control what AI agents do on other systems, but it can make sure that on your Mac nothing happens without your explicit permission.
What you can do
- ✓Check which apps have Full Disk Access in System Settings.
- ✓Revoke the permission from apps that do not need full access.
- ✓Update macOS as soon as Apple releases the new privacy controls.
How many apps have full access to my disk without my knowledge?
Security is not improvised, it is audited. At Nacata Security we detect vulnerabilities and protect your company, because a single flaw can cost you everything you have built.
Related articles
Nacata Security, reach out to us anytime
We are Nacata Security, get to know us
web: nacata.io
email: info@nacata.io
Phone: 919930793
LinkedIn: Nacata Security
The intersection of artificial intelligence and cybersecurity: attacks that use AI to deceive or impersonate, security of models and conversational assistants, and new AI-based threat detection tools.
RATING
9.7
Who are we?
At Nacata Security we are an offensive cybersecurity company specialized in audits and penetration testing.
We detect, assess and help mitigate the vulnerabilities of your systems, networks and applications before a real attacker exploits them, offering 360º defense tailored to each client.
We’d be glad to get in touch with you for whatever you need.




