New TEE.fail Exploit Steals Secrets from Intel & AMD DDR5 Trusted Environments

Researchers have unveiled a new attack called “TEE[.]fail” that undermines the security guarantees of Intel and AMD Trusted Execution Environments (TEEs) by targeting DDR5 memory. According to GBHackers on Security, the attack uses a memory interposition device that physically intercepts DDR5 memory traffic in server environments, and can be built with off‑the‑shelf components, making it economical and practical for determined adversaries.

The method exploits how Intel TDX and AMD SEV-SNP encrypt data in memory, which the researchers describe as using deterministic encryption that can produce identical ciphertext for identical plaintext blocks, enabling a form of comparison attack without decrypting content. In a proof‑of‑concept, a DDR5 DRAM bus interposer captured memory transactions in real time, with three write operations—zeros, ones, zeros—producing identical encrypted values for the matching operations.

The team also demonstrated extraction of ECDSA attestation keys from Intel’s Provisioning Certification Enclave, forging TDX quotes that pass verification at the highest “UpToDate” trust level, thereby compromising the attestation chain. According to the report, the attack’s portability extends to Nvidia’s GPU Confidential Computing and could enable illicit activity such as unauthorised AI workloads, underscoring urgent datacenter security implications.