I guess something like that: https://tpm2-software.github.io/tpm2-tss/getting-started/201...

Basically TPM includes key that's also signed with manufacturer key. You can't just extract it and signature ensures that this key is "trusted". When asked, TPM will return boot chain (including bootloader or UKI hash), signed by its own key which you can present to remote party. The whole protocol is more complicated and includes challenge.

Tpm isn't designed for this use case. You can use it for disk encryption or for identity attestation but step 1 for id attestation is asking the tpm to generate a key and then trusting that fingerprint from then on after doing a test sign with a binary blob. The running kernel is just a binary that can be hashed and whitelisted by a user space application. Don't need tpm for that.

This is called the Endorsement Key, and you're correct, it never leaves the TPM. The TPM is a "black box" to the OS.

Ah, got it. With enough motivation this is still pretty easily defeated though. The key is in some kind of NVRAM, which can be read with specialized equipment, and once it's out, you can use it to spoof signatures on a different machine and cheat as usual. The TPM implementations of a lot of consumer hardware is also rather questionable.

These attestation methods would probably work well enough if you pin a specific key like for a hardened anti-evil-maid setup in a colo, but I doubt it'd work if it trusts a large number of vendor keys by default.

Once it's out you could but EKs are unique and tied to hardware. Using an EK to sign a boot state on hardware that doesn't match is a flag to an anti-cheat tool, and would only ever work for one person.

It also means that if you do get banned for any reason (obvious cheating) they then ban the EK and you need to go source more hardware.

It's not perfect but it raises the bar significantly for cheaters to the point that they don't bother.

> Using an EK to sign a boot state on hardware that doesn't match is a flag to an anti-cheat tool

The idea is you implement a fake driver to sign whatever message you want and totally faking your hardware list too. As long as they are relatively similar models I doubt there's a good way to tell.

Yeah, I think there are much easier ways to cheat at this point, like robotics/special hardware, so it probably does raise the bar.

Any sane scheme would whitelist TPM implementations. Anyway fTPMs are a thing now which would ultimately tie the underlying security of the anticheat to the CPU manufacturer.