TLBleed is a cryptographic side-channel attack that uses machine learning to exploit a timing side-channel via the translation look-aside buffer (TLB) on modern microprocessors that use simultaneous multithreading.[1], the attack has only been demonstrated experimentally on Intel processors; it is speculated that other processors may also potentially be vulnerable to a variant of the attack, but no proof of concept has been demonstrated.[2] AMD had indicated that their processors would not be vulnerable to this attack.[3]
The attack led to the OpenBSD project disabling simultaneous multithreading on Intel microprocessors.[4] [5] The OpenBSD project leader Theo de Raadt has stated that, while the attack could theoretically be addressed by preventing tasks with different security contexts from sharing physical cores, such a fix is currently impractical because of the complexity of the problem.[4]