DDRop hardware attack bypasses Intel and AMD memory encryption

Researchers from KU Leuven, ETH Zurich, Durham University, and Google have identified a hardware vulnerability named ‘DDRop’ that can bypass memory protections in modern confidential computing environments. The attack targets Intel TDX, Intel Scalable SGX, and AMD SEV-SNP technologies, which are designed to protect data during processing in cloud servers.

The vulnerability stems from a design flaw where scalable memory encryption hardware fails to verify ‘freshness’—the guarantee that the data being read is the most recent version written. By using a custom-built hardware interposer costing less than $200, attackers can sit between the processor and the RAM module to interfere with DDR5 write operations. By corrupting commands on the high-speed memory bus, the device can silently drop specific writes, causing the protected virtual machine to continue computing using old, stale data that still decrypts correctly.

While the attack is effective, it requires physical access to the server to install the interposer. The researchers have released the complete interposer design as open-source hardware to facilitate further study.