DDRop Attack against Intel TDX and AMD SEV-SNP

Technical Analysis

DDRop targets the integrity guarantees of Trusted Execution Environments (TEEs) by manipulating memory write paths at the hardware/DRAM interface level, causing write operations to be silently discarded while the CPU's encrypted memory controller continues serving previously cached ciphertext as valid current state. This breaks the confidentiality and integrity assumptions of Intel TDX and AMD SEV-SNP, which are designed to protect VM memory even from a compromised hypervisor or cloud provider insider. The attack requires physical proximity and pre-existing software access to insert a small interposer circuit, placing it in the category of sophisticated supply-chain/physical hardware attacks rather than remote exploitation, though it undermines a foundational trust boundary for cloud-hosted confidential workloads. No CVE has been assigned as of this report; affected vendors will likely need microcode, memory controller, or attestation-policy mitigations. Organizations running AI agent infrastructure, RAG pipelines, or LLM inference workloads inside confidential VMs (e.g., for protecting model weights, API keys, or sensitive prompt/context data in multi-tenant cloud environments) should treat TDX/SEV-SNP attestation as potentially bypassable in threat models that include a malicious insider or compromised hardware supply chain, since stale-data injection could corrupt agent state, poison retrieved context, or leak credentials assumed to be hardware-isolated.

Affected Systems

Servers using Intel Trust Domain Extensions (TDX); Servers using AMD Secure Encrypted Virtualization-Secure Nested Paging (SEV-SNP); Confidential computing cloud instances built on these technologies

Indicators of Compromise

• Not applicable — this is a hardware-level physical attack technique, not a malware/network-based threat with file, hash, or network IOCs

Remediation Steps

1. 1

   Review confidential computing threat model

   Update risk assessments for TDX/SEV-SNP deployments to account for physical/hardware tampering scenarios, especially in shared or co-located hardware environments.

2. 2

   Restrict physical and privileged access

   Tighten physical security controls and privileged software access to servers hosting confidential VMs to reduce the window for hardware implant insertion.

3. 3

   Monitor vendor advisories

   Track Intel and AMD security advisories for microcode updates, attestation improvements, or mitigations addressing DDRop once available.

4. 4

   Reassess sensitive workload placement

   For AI agent, LLM, or key-management workloads relying on TEEs for isolation, evaluate additional layered protections (e.g., HSMs, remote attestation with supply-chain verification) rather than relying solely on TDX/SEV-SNP guarantees.

5. 5

   Engage cloud providers

   Request clarification from cloud/hosting providers on physical security controls and hardware supply-chain integrity verification for confidential computing offerings.