Anssi: Technical Position Paper on Confidential Computing (cyber.gouv.fr)

France’s cyber‑security agency ANSSI has published a technical position paper on confidential computing that frames how hardware‑backed Trusted Execution Environments (TEEs) should be used, evaluated and deployed. The paper defines confidential computing as protecting data “in use” (not just at rest or in transit), surveys common technologies (Intel SGX, AMD SEV, ARM TrustZone and cloud enclave offerings), and sets out a risk‑based lifecycle: threat modelling, attestation and key management, secure provisioning, patch/firmware management, logging and controlled decommissioning. This guidance matters for AI/ML teams that increasingly rely on cloud GPUs and third‑party inference: TEEs can protect model IP and sensitive training or inference data, but ANSSI warns they are not a panacea. Key technical takeaways include the need to validate remote attestation chains, minimise the Trusted Computing Base, mitigate side‑channel and microcode/firmware vulnerabilities, and integrate confidential computing with broader supply‑chain, identity and cryptographic controls. ANSSI also pushes for vendor transparency, standardised attestation and certification paths so organisations can make informed tradeoffs between performance, security and auditability when protecting AI workloads.