The Scope Creep Automated reviews of new AI-related working group submissions

Review · August 21, 2026

WIMSE Workload Attestation Summary

draft-reddy-wimse-workload-attestation -00View on Datatracker →

Problem Statement

WIMSE's existing workload-to-workload authentication (WIT + WPT) proves who a caller is and that it holds the right private key, but says nothing about where the workload is running or whether that key is hardware-protected. Existing TLS-layer attestation mechanisms (from RATS/SEAT) bind attestation cryptographically to the TLS connection — which breaks as soon as a TLS-terminating proxy is in the path. This draft solves the problem by carrying attestation evidence as HTTP header fields alongside the existing WIT/WPT headers, preserving end-to-end security without requiring proxy modifications.

How the Spec Fits In

This is a targeted extension to WIMSE workload-to-workload authentication. It introduces two new HTTP header fields and defines protocol flows for two existing RATS models (background check and passport). It creates no new roles within OAuth but introduces the RATS Verifier role as a new participant in WIMSE flows.

The spec depends on a significant number of other documents:

  • The WIMSE core specs (WIT, WPT)
  • RFC 9334 (RATS Architecture) and its background-check and passport models
  • CMW (Conceptual Message Wrapper) for Evidence encapsulation
  • EAT (Entity Attestation Token) and EAR (EAT Attestation Result)

This is a leaf-node spec that integrates existing building blocks rather than defining new cryptographic primitives. It is limited in scope to the HTTP transport layer for WIMSE.

High-Level Summary

The spec adds two optional HTTP header fields that a workload can include alongside its existing WIT and WPT authentication headers:

  • Workload-Evidence — raw TEE-generated attestation evidence (e.g., from a TPM or hardware enclave), wrapped in a CMW container that identifies the attestation technology used.
  • Workload-Attestation-Result — a pre-verified attestation result (EAR) from a trusted Verifier, for use when the workload has already been appraised before making the request.

The common thread across all three headers is the workload's public key: the WIT asserts it as the workload's identity key, the WPT proves possession of the corresponding private key, and the attestation evidence confirms that the key is hardware-protected in a trusted execution environment.

Binding is achieved by using the WPT's unique token ID (jti) as the nonce in the attestation evidence, ensuring that a given attestation result cannot be replayed with a different request. The backend (Relying Party) verifies all three headers and must confirm that the public key across all three matches before accepting the request.

Two flows are supported:

  • Background check model: The workload sends raw evidence with each request; the backend calls out to a Verifier to appraise it.
  • Passport model: The workload pre-appraises its own evidence with a Verifier before making each request and carries the result as a signed EAR.

Interoperability

Strong Points

The core binding mechanism — using the WPT jti as the attestation nonce — is well-specified and critical for preventing replay attacks. The verification algorithms for both models are detailed step-by-step, which should lead to consistent implementations. The mutual exclusivity rule (both headers present → 400) avoids ambiguity.

Significant Interoperability Concerns

The spec relies on CMW for Evidence encapsulation, which abstracts over a wide variety of TEE formats (TPM, Intel TDX, ARM CCA, etc.). There is no requirement to support any specific Evidence format, so in practice two implementations may both be "conformant" while being unable to interoperate because they support different attestation technologies. This gap is not addressed.

Key format comparison is a likely source of implementation friction: the EAR carries the verified key in PEM-encoded SPKI or as a certificate, while the WIT carries it as a JWK. The spec acknowledges that format conversion is required but leaves the mechanics entirely to the implementer. Small mismatches in how implementations handle this conversion (e.g., point compression, curve representation) could cause interop failures that are difficult to diagnose.

Mechanisms That Could Be Removed

The spec states that proxies must forward both header fields unchanged. This is already the default HTTP behavior for unknown headers — the normative MUST adds no new constraint that would affect interoperability and could be dropped or relegated to a note.

Writing Quality

The writing is generally clear and uses appropriate IETF language. The use of a summary table showing how the three header fields relate to one another is a helpful addition. The verification algorithm sections are structured well and should be actionable for implementers.

There are a few areas that could be improved:

  • The spec says the EAR's ear_verified_attester_key can be either a PEM SPKI or a certificate but does not specify which is preferred or what validation is required for the certificate case. This ambiguity should be resolved.
  • The term "public-key confirmation claim" in the Workload-Evidence section is used without a precise reference to where this claim is defined for each supported attestation format. Implementers will need to know the exact claim name or structure per format.
  • The applicability section states that both header fields work with both WIMSE protection mechanisms (WPT and HTTP Signatures) but does not explain how the nonce binding works in the HTTP Signatures case (where there is no jti equivalent to reference).

Overall the spec is appropriately concise for what it describes.

Author Background

Tirumaleswar Reddy (Nokia) is a very prolific IETF participant with 39 published RFCs and 37 active Internet-Drafts across a wide range of areas including TLS, DNS, IPsec, JOSE/COSE, and post-quantum cryptography. He has Security Area Directorate and Ops Directorate reviewer roles. He is one of the most active individual contributors currently writing in the IETF.

Nathanael Ritz (Independent) is a newer IETF author with 0 published RFCs and 6 active drafts. His drafts appear concentrated in the RATS, WIMSE, and SEAT areas, suggesting focused expertise in workload attestation. He is not yet an established RFC author but is actively contributing in this specific domain.