DDR5 interposer reportedly enables replay attacks on confidential VMs

Dropping writes to keep stale data in use

Security researchers described DDRop, a small circuit board placed between a processor and a DDR5 memory module. They say it can silently drop writes, causing a protected virtual machine (VM) to continue using older data that still decrypts.

The researchers reported a proof of concept against an Intel TDX server, claiming that crafted page-table entries could put a protected VM into debug mode and expose plaintext private memory. They also said altered TDX metadata could produce forged attestation reports. The attack requires physical access to the target server.

According to the researchers, DDR5 had previously seen only weaker passive attacks, while a similar active interposer attack had been demonstrated on DDR4.

The reported DDRop design moves the attack to DDR5 with an active, under-$200 interposer that alters memory-bus traffic at full speed.

The physical-access boundary

Confidential-computing systems are intended to protect data even while it is being processed. The reported attack targets memory freshness—the ability to establish that data is current—rather than a software vulnerability. Intel and AMD reportedly regard this physical-access scenario as outside their cloud threat models; the researchers say current scalable memory-encryption designs lack a simple root-cause patch.