Researchers expose cheap hardware flaw that compromises encrypted cloud memory

Researchers from KU Leuven, ETH Zurich, Durham University and Google discovered a design weakness in scalable memory-encryption hardware that does not check whether stored data is fresh. They engineered an inexpensive DDR5 interposer, called DDRop, which sits on the memory bus and silently suppresses write operations, forcing a protected virtual machine to continue processing outdated ciphertext that still decrypts correctly.

In tests on an Intel TDX server, the technique allowed the attackers to place the VM into debug mode, read its memory in plaintext and forge attestation reports, all deterministically within two minutes and without crashing the system. The method works against Intel TDX, Scalable SGX and AMD SEV-SNP, and differs from prior passive attacks by operating at full bus speed. Intel and AMD have acknowledged the vulnerability but consider it out of scope for their cloud security models, and no software or hardware patch is currently available. The researchers released the interposer design as open-source hardware.