A Cheap Interposer Against Expensive Trust Guarantees
Researchers have disclosed DDRop, a new hardware attack that breaks the memory-protection guarantees of Intel and AMD confidential computing platforms by silently dropping writes to a server's memory — causing the processor to keep reading old encrypted data as if it were still current.
The attack uses a small circuit board called an interposer, costing under $200, inserted between a processor and a memory module. It forces errors on the memory command bus to drop writes, then disconnects the error-reporting wire so the processor never learns the write failed. Because the affected encryption schemes lack freshness checks, the processor has no way to detect that it's reading stale data.
Affected Technologies
| Technology | Impact |
|---|---|
| Intel TDX | Full virtual machine compromise — reading victim memory, enabling debug mode, and forging attestation measurements |
| AMD SEV-SNP | Copying victim page contents via the page-relocation feature |
| Intel Scalable SGX | Also vulnerable to the same class of write-dropping attack |
| Intel Client SGX (older) | Not affected — different architectural protections |
| NVIDIA GPUs | Not affected — different architectural protections |
Attack Requirements
DDRop is not a remote, network-based exploit. An attacker needs:
- Control of server software running on the target
- Brief physical access to the hardware to insert the interposer
- The ability to operate the interposer purely through software once installed
This puts the attack squarely in the threat model of malicious cloud insiders, supply-chain tampering, and physical data-center compromise — precisely the scenarios confidential computing was designed to defend against.
Who Found It
The research comes from a collaborative team at KU Leuven, ETH Zurich, Durham University, and Google. Findings will be presented at ACM CCS 2026 in November.
Vendor Response
Intel and AMD acknowledged the findings through coordinated disclosure but have not published mitigation guidance or a timeline. Both vendors maintain that physical interposer attacks fall outside their published threat models, and Intel has stated it will not assign a CVE to this class of attack. Intel's optional cryptographic-integrity mode offers partial mitigation against the technique.
Why This Matters
Confidential computing is the foundation many cloud providers market as protection against a compromised hypervisor or a malicious data-center operator. DDRop demonstrates that the memory-encryption schemes underpinning Intel TDX and AMD SEV-SNP still trust the memory bus to behave honestly — an assumption that a cheap physical implant can violate. Organizations relying on confidential computing purely for regulatory or contractual assurance should treat physical security of the underlying hardware as an equally critical control.