Apple will update Macs to protect users from AI agents with full disk access
7 days ago by TryingToBeGood to c/technology
It's not clear exactly what Apple is going to impose here. In principle, trying to ensure users understand the implications of a permission is good. In practice, Apple is famous for keeping users from making decisions they might reasonably want to make on phones, and it's starting to creep into the Mac; installing an app Apple hasn't signed requires the command line, for example.
Apple also has a history of requiring excessively scary permission requests for things that Apple apps do without prompting. They abuse the permission system to foster a sense that Apple products are safe and any competitor is dangerous.
Competitors meaning also third party apps.
Apple's apps have inherently more permissions than third party apps can ever dream of
Even from their store. They were being investigated by German antitrust authorities and last month they agreed to reduce the fear mongering for EU users. The agreement doesn't apply to the rest of the world.
installing an app Apple hasn't signed requires the command line, for example
It doesn’t, but it does make a scary popup, and requires the user to know to go to settings to find the section to override that popup.
installing an app Apple hasn’t signed requires the command line, for example
Yeah this is the reason I've avoided Macs for so long. I get giving the user a warning but not providing any sort of prompt to override the warning, and hiding it in a terminal command is ridiculous .
There was also one time I tried to simply turn off Bluetooth and it was just like "I'm sorry Dave, I can't do that". And I was like WTF, absolutely not .
Yeah I love meta saying “it’s impossible if you’ve denied it access.”
If the user gave it a task and it decided the easiest path forward was just to read his messages, that’s no different than OpenAI agents breaking protocol to find the fastest path to their goal, regardless of side effects (assuming the “permission” was just a request from the LLM rather than a request for full disk access in general from the meta app).
I think Apple is shining light on an important reality here:
In the modern era, the populace is practically required to use technological devices that it is unreasonable to expect them to fully understand. So, rather than leave the consequences of that lack of understanding to fall on them with a shrug, it is better for everyone if those devices and their makers do their best to help them by limiting the impact of their ignorance of that particular technological niche.
In that vein, I feel that it is long overdue to apply "the principle of least privilege" and a "defense in depth" philosophy to home computing devices and operating systems. Compartmentalizing data and functionality behind fine-grained permissions with verbose, plain speech descriptions of what those permissions are for, along with what should and should not be requesting those permissions, would go a long way to prevent these sorts of data abuses. It would also help to limit how much damage viruses and malware could do if installed.
For example. I'm frankly struggling to come up with any valid use cases for full disk access on a computer beyond file backups. I'm sure there are some, but almost nobody would need to use them, so a dire, flashing warning seems in order.
It is clear that sane regulation of unscrupulous software behavior won't be forthcoming any time soon, but even then, you still should lock your doors even though trespassing is illegal, metaphically speaking. "Trust but verify," one of the only good statements out of Ronald Reagan's mouth.
The apps i use have to request disk access which i can approve or not. I always assumed that meant problems like this were not possible. I think many of us have been under a false impression because of this. I assumed the other apps were secured, which was foolish of me, I admit.
I don't think you have to be so critical of yourself. Everyone here uses things every day that they have incomplete or maybe even even no understanding of how parts of it work. Its just a reality, maybe an unfortunate one, of the level of technological advancement we're at and how we've incorporated it into our societies.
I think the issue is that it becomes an everything or nothing. some people want to give apps access to their own documents but don't realize their are also giving access to things like their chat messages, cookies, etc.
If they aim to provide a more granular approach while still allowing full access to those that need it, Im ok with that
All of computing beyond programming with switches is “protecting the user from themselves.”
“Why use a language? Aren’t you able to perfectly program in binary? I’m not going to waste time building something just because you’re stupid.”
I read the article, and I think DickHertz's comment is pretty much right. The article's example illustrates the point:
For example, Inc. columnist Jason Aten recently discovered that Meta's Muse AI assistant somehow had synced his entire local Messages database and was using those messages as context for its tasks.
Aten did not understand that the Messages database is stored on the disk (presumably) unencrypted and readable by anything with full disk access. He got a result he didn't want because he did not understand the implications of granting that permission, and Apple seeks to add more friction to the process to protect users from making that mistake.
Even if it was encrypted, it could be unlocked because with full disk access it can read the local private key, which probably doesn’t need a passphrase because it’s be a pain in the ass for the user to retype that every reboot or more
Starting to see why you shouldn’t let any corporate AI loose on your daily driver yet?
Macs have a hardware secure enclave that apps can use to store keys where other apps can't access them. Apple itself does not seem to be making adequate use of it here.
Yeah, I read it. That’s kind of the point. Apple’s adding more guardrails because people will click Allow on damn near anything without thinking about what they’re actually giving access to. 🙄
That really depends on how you design the pop-up dialogue. If it looks like a standard EULA, yeah, most people will just skip reading. If you darken the screen, and use large, flashing warnings with a short and clear info about the limited amount of cases for such a dangerous permission to be ok, they will for sure look closer.
Late Edit: And make sure you don't fatigue your clients with a permanent spam of warnings! Holy fuck, MS with their UAC really had to learn that one the hard way. If every single software installation and change of system settings warns you of proceeding then the warning becomes useless (and these warnings were actually one of the first things i deactivated on fresh installations exactly of that issue)
Wish i could integrate it with my Bluetooth siren and a configurable sci-fi one-liner: “Danger, Will Robinson!” “Just what do you think you’re doing, Dave?” “Nuh uh uhh” etc
It’s also because Apple’s permissions are hilariously monolithic. A lot of their permissions are set up as an all-or-nothing event, instead of allowing granular control. Full disc access is a great example, where a user should be able to grant access to specific folders and files on an as-needed basis. But that’s not the default behavior. The default behavior is to just go “hey do you wanna give this app access to everything?
Or an update adds it without your knowledge…
Apps can’t add permissions via an update. If an app adds a new feature with an update that requires a new permission, the user will be prompted the first time the app tries to use that new feature.
An "open" platform?
Edit: Of course, here come the fanboys lol it's ok. I was one of you once. You'll grow.
Being able to run a program on an OS is the bare minimum. Anything not signed by Daddy Apple, you have to add an exception. The moment the exceptions are no longer enough to "protect the user against themselves," you can't do a thing. Why? Because it's a fully closed, walled platform.
It's like being in prison and being told you're free because you can walk in the yard.
The Source?
If we’re ever going to have a proper framework for governing agents, it will include utilities in most abstraction layers (to include, importantly so, the OS).
I’ve had an interesting idea on using TreeOS to help make something a little more natively AI-agent safe. It’s a project on my back burner right now. Something about making changes branch-able and version controlled by default, copy on write by default, … I need to really think it through more.
I was thinking an agent should be like a user too. A first-class entity to be managed, but probably ephemeral in nature. Not a static user. Probably more like a permissions set that arbitrary agents can be associated with.
@lemmy.world
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
go to feed...
@lemmy.world
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
go to feed...
Good change.
Software should not operate at a privileged level “just in case”, but rather have a well-defined functional boundary.
save