Comment by red_trumpet

DRM doesn't protect, it only delays.

Never trust the client. Anything the client has access to, whether "protected" by Play Integrity or not, should be considered compromised.

It makes sense, to some degree, that for example some banking apps refuse to run if they detect that the phone has been rooted, or even to go as far as to refuse to run if there are non-Play apps on the phone.

Maybe some apps with DRM media playback do this kind of check too, yes. Haven’t used Android for many years now.

Hopefully iOS stays the way it is where apps don’t get so much info about other apps on device. I prefer it that way.

What's implemented wrongly is the idea that this kind of client side check somehow removes the need for server side verification.

And the idea that this kind of check can't be defeated.

GPIDRM doesn't protect against much, even if it's perfect [0]. What it gives you is an API your Android app can call into to inspect the device status.

That's not enough because the owner of the phone can just twiddle that memory between you calling the API and using the value. You fully own the code that runs on your devices, and if you don't like it then you can just choose to run different code. The GPIDRM hinders some users who want to fully own their device and also use your app, but it doesn't actually protect your app from being executed in other environments (similarly with any other modification to how the GPIDRM might function, short of it physically decrypting the code/data you intend to run and only ever running in environments that would somehow prevent people from backing up those decrypted bytes -- or, similarly, physically decrypting data unique to a particular instance of using the app and not useful for any reason when somebody else runs the app).

When, then, does GPIDRM make sense to use?

_Arguably_ the thing that banks do isn't terrible [1]. Their servers are authenticated, so it's not a security thing. They're just managing risk (people with rooted phones might be more likely to have root-level malware for example). If somebody has a rootkit leaking banking details and the attacker is also willing to pay $10 to borrow their phone number for the day, the bank account will be fully compromised. When that happens, the bank is on the hook some fraction of the time. The bank server trusts requests to either come from a real user or a user with stolen credentials, and they're trying to reduce the chance of the latter (but not eliminate, even from rooted Android phones).

How does McDonald's differ? There are no server-side checks, no passwords, no logins, no crypto handshakes, no anything. If you send a request pinky promising you're a trusted client then you'll get your free food. The implementation was so bad that the TFA demonstrated compromising it on a phone which _correctly_ passed the GPIDRM check.

[0] No such technique can be perfect. At its core, it relies on a secure hardware enclave. Physical keys are always reversible with enough time and effort, in time _linear_ in the key length. The goal is just to create a constant factor big enough that almost nobody with expensive enough tools to dismantle the chip and go probing is willing to go through the effort (or, ideally, not able to with the current generation of technology, so that rotating keys every few years can keep up with reversing efforts).

[1] I'd be shocked if people with rooted Android phones were actually more likely to be victims of phishing/malware/....