Review · August 21, 2026
Transitive Attestation for Sovereign Workloads — A WIMSE Profile
Problem Statement
Bearer tokens and private keys are "context-agnostic" — a stolen credential works from any environment. This spec introduces a mechanism called Proof of Residency (PoR) to cryptographically bind a workload's identity to its local execution environment, making credentials effectively "sticky" to the place where they were issued. The stated motivation is amplified by AI agent scenarios where prompt injection attacks could exfiltrate credentials for reuse elsewhere, and by regulatory requirements that workloads stay within specific jurisdictions.
How the Spec Fits In
This is a WIMSE profile that assembles pieces from several working groups into a new composite mechanism:
- It extends DPoP (RFC 9449) with a new parallel construct called DPoR (Demonstrating Proof of Residency), binding key possession proofs to a local workload agent's signature.
- It uses TLS Exporter (RFC 5705) as a channel-binding primitive to tie residency proofs to specific TLS sessions.
- It depends on RATS concepts (hardware attestation, TPM/TEE quotes) for its high-assurance tier.
- It references SPICE for delegation chain handling (east-west identity flow).
- It explicitly defers platform attestation and location evidence to a companion "Verifiable Geofencing" profile, which does not yet appear to exist as a published draft.
The spec creates a new role — the Workload Identity Agent (e.g., SPIRE Agent) — that sits between the workload and the verifier, signing residency proofs on behalf of the workload via a local channel (Unix Domain Socket). This is a new relationship not defined in core WIMSE or OAuth.
High-Level Summary
The spec defines two mechanisms for proving workload residency:
mTLS-Based Transitive Attestation: After a mutual TLS handshake, the workload asks a local identity agent (running on the same host) to sign a payload that includes a hash of the TLS session's Exporter value and a server-provided nonce. This signed "Proof of Residency" is sent to the resource server, which can verify that the entity presenting the mTLS certificate is the same entity that obtained the local agent's signature — confirming both who the workload is and where it is running.
DPoR (Demonstrating Proof of Residency): An extension of DPoP where the same concept applies to bearer-token flows. The workload's local agent signs a payload binding the DPoP key thumbprint to a server-provided nonce, proving that the DPoP key is co-located with the trusted agent — not exported to a foreign environment.
For high-assurance deployments, the local agent runs inside a hardware Trusted Execution Environment (Intel TDX/SGX), becoming a "logical quoting enclave" that bridges hardware-rooted trust to workload-level identities. The TLS Exporter hash is embedded in the TEE attestation report, closing the loop: the verifier can confirm the TLS peer is the exact entity that generated the hardware attestation.
Re-attestation is tiered: lightweight identity renewal happens frequently (using cached platform evidence), while heavyweight hardware re-attestation happens on a slower schedule.
Interoperability
Significant interoperability concerns: The two assurance tiers — standard co-location vs. high-assurance hardware-rooted — are likely to be implemented inconsistently across deployments. A verifier accepting either tier cannot know in advance which tier a given client will present, and the spec does not define how tier negotiation works. In practice this may lead to deployments where one side implements only the standard tier and the other requires the high-assurance tier, with no defined fallback path.
The local agent communication channel (Unix Domain Socket) is described conceptually but the wire protocol is not specified. Different SVID/SPIFFE implementations may behave differently here, undermining interoperability at the most critical step.
The security considerations section has multiple explicit TBD gaps: nonce entropy requirements, clock skew handling, and agent compromise scenarios. These are not minor — nonce entropy and clock skew are directly load-bearing for the security model.
Mechanisms that could be removed or deferred: The confidential computing section (4.3) describes a detailed hardware attestation architecture that overlaps with RATS work. Given the spec already defers platform attestation to the companion "Verifiable Geofencing" profile, this section may be redundant or premature while that companion document does not exist. Removing it would substantially shrink the scope to the interoperable core (DPoR and mTLS binding), which is the more novel contribution.
The "Tiered Attestation Schedule" and "Fast Path renewal" concepts add complexity without sufficient protocol-level specification to be interoperable.
Writing Quality
The writing mixes IETF-style protocol description with marketing language that is unusual for a standards document. Terms like "Silicon-to-SVID chain of accountability," "Sovereignty Violation," "Workload Fusion Nonce," and "North-South security axis" are invented vocabulary that doesn't correspond to established IETF terminology. While some of these are defined within the document, the use of proprietary-sounding coinages makes the spec harder to evaluate and review.
The protocol flows in sections 4.1 and 4.2 are reasonably clear and follow the numbered-step format common in IETF specs. However, the spec relies heavily on SPIRE/SPIFFE-specific concepts (SVIDs, Workload API) without sufficiently abstracting them, giving the impression of specifying an implementation rather than an interoperable standard.
Several critical details needed for implementation are deferred to TBDs, and the dependency on a companion "Verifiable Geofencing" profile that does not appear to exist makes portions of the spec incomplete. The spec is at version -00, so these are not disqualifying, but significant work remains to reach the quality bar for adoption.
Authors
Ram Krishnan (JPMorgan Chase, [email protected]) is a seasoned IETF contributor with 6 published RFCs dating back to 2002, including RFC 3270 (MPLS Differentiated Services, cited by 50 RFCs). He has 7 currently active drafts. His recent work has shifted from networking/virtualization toward security and attestation topics, including several drafts closely related to this one.
A. Prasad (Oracle, [email protected]) has no published RFCs and is a relatively new IETF author. However, they have 7 active drafts spanning OAuth, SPICE, RATS, and WIMSE — suggesting rapid and productive engagement across multiple working groups simultaneously.
Diego R. Lopez (Telefonica, [email protected]) is a very experienced IETF contributor with 12 published RFCs (2016–2023), including RFC 8253 (PCEPS/TLS, cited by 45 RFCs). He has 22 active drafts and currently chairs the IETF GREEN working group. His breadth of involvement is notable, though his primary expertise appears to be in network management and operations rather than identity/security.
Srinivasa Addepalli (Aryaka, [email protected]) has no published RFCs and is a newer IETF participant. Like Prasad, they have 7 active drafts covering overlapping topics (attestation, OAuth, SPICE), suggesting this group of authors is working together on a cluster of related proposals.
The author group presents an interesting mix: two highly experienced IETF contributors lending credibility, alongside two newer contributors who appear to be the primary drivers of this cluster of related drafts. The concentration of related active drafts across the same small group of authors warrants attention — several companion drafts referenced in this spec are authored by the same people, which could indicate a coordinated push across working groups for a particular architectural vision.