This is a great list of academic attacks, but it proves less than you think.
Yes, TEEs have been broken in dozens of ways. Side channels, transient execution, voltage manipulation, interrupt timing... etc. To be fair, you could make an equally impressive list for many security primitives.
The question isn't "can TEEs be broken?" since clearly they can, but rather what's your threat model and what are your alternatives?
What TEEs actually defend against is passive compromise. They force an attacker to actively exploit rather than just read memory. That legal and operational distinction matters enormously in practice.
The alternative to TEE is "no hardware isolation at all," and that's strictly worse for every threat model where TEEs provide value.
Additionally, you still get attestation which gives you cryptographic proof of what code is running.
2 comments
[ 3.3 ms ] story [ 17.0 ms ] threadYes, TEEs have been broken in dozens of ways. Side channels, transient execution, voltage manipulation, interrupt timing... etc. To be fair, you could make an equally impressive list for many security primitives.
The question isn't "can TEEs be broken?" since clearly they can, but rather what's your threat model and what are your alternatives?
What TEEs actually defend against is passive compromise. They force an attacker to actively exploit rather than just read memory. That legal and operational distinction matters enormously in practice.
The alternative to TEE is "no hardware isolation at all," and that's strictly worse for every threat model where TEEs provide value.
Additionally, you still get attestation which gives you cryptographic proof of what code is running.