]> Pre-, Intra- and Post-handshake Attestation TU Dresden
muhammad_usama.sardar@tu-dresden.de
Security Secure Evidence and Attestation Transport remote attestation TLS 1.3 attested TLS This document presents a taxonomy of extending TLS protocol with remote attestation, referred to as attested TLS. It also presents high-level analysis of benefits and limitations of each category, namely pre-handshake attestation, intra-handshake attestation and post-handshake attestation. It also captures the opinions of the WG participants in order to build consensus towards solutions. It also discussed tradeoffs and scalability. About This Document The latest revision of this draft can be found at . Status information for this document may be found at . Discussion of this document takes place on the Secure Evidence and Attestation Transport Working Group mailing list (), which is archived at . Subscribe at . Source for this draft and an issue tracker can be found at .
Introduction Based on our extensive analysis of attested TLS , we classify attested TLS into three main categories:
  • pre-handshake attestation,
  • intra-handshake attestation, and
  • post-handshake attestation.
In pre-handshake attestation, the signing of Claims precedes the TLS handshake, while post-handshake attestation applies the reverse. Intra-handshake attestation requires the signing of Claims to be done within the TLS handshake protocol.
Scope In this version, we analyze the three categories (without combinations) with a focus on the last two, i.e., intra-handshake attestation and post-handshake attestation. The current scope of this draft is existing specifications and real-world implementations pointed in the given references. Any theoretical solutions are currently out of scope until some specification or implementation emerges. For simplicity, we consider simple Attester with only one Attesting Environment and only one Target Environment . That is, complicated scenarios such as Composite Device etc. are out of scope in this version. From RATS perspective, we consider Background Check Model . Future versions will add Passport Model . From TLS perspective, the scope is limited to TLS 1.3 as per . That is, older versions of TLS are explicitly out of scope.
Note Regarding remote attestation, we note that:
Remote attestation provides guarantees about the state of Attester only at the time at which signing of Claims is done to generate Evidence .
Conventions and Definitions The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 when, and only when, they appear in all capitals, as shown here. We use terminology from and slightly loosely (intentionally) for readability. Future versions will tighten it. In addition, we define three temporal terms:
  • Evidence Generation Time: Time when Evidence is generated (more specifically when Claims are signed)
  • Connection Establishment Time: Time at which TLS handshake is performed
  • Lifetime of Connection: Time period starting from Connection Establishment Time until the connection exists.
Pre-handshake Attestation
Benefits In certain deployments, there may be benefits for pre-handshake attestation. As Yaroslav Rosomakho (as individual contributor) observes :
no modification of TLS, no changes to application protocols, and potential caching/scalability.
He (as individual contributor) further explains :
Many signed artifacts carry issuance time and validity constraints (e.g., JWT iat/exp , HTTP Message Signatures created/expires , SAML conditions ). A timestamp alone is not the same as freshness, but a short validity window plus a trustworthy time source can be an acceptable assurance mechanism in some deployments. Likewise, some schemes can provide stronger freshness via verifier-chosen challenges or monotonic counters.
Limitations There are also security concerns on pre-handshake attestation for use cases such as confidential computing . Since the Evidence Generation Time could be at any arbitrary point of time in the past compared to the Connection Establishment Time in the case of untrusted clocks as in confidential computing, pre-handshake attestation provides no guarantees about the state of Attester at the Connection Establishment Time and during the Lifetime of Connection.
Intra-handshake Attestation In general, intra-handshake attestation improves the situation where Evidence Generation Time is the same as Connection Establishment Time, assuming freshness mechanisms, such as unpredictable, single-use challenge and clear replay handling, are in place. In following subsections, we present the benefits and limitations of intra-handshake attestation.
Benefits
No Additional Application-Level Protocol Intra-handshake attestation does not require a new application-layer protocol or message exchange. Evidence and related metadata are conveyed within handshake via TLS extensions. TLS is responsible for conveyance of the Evidence; it does not perform appraisal of Evidence or authorization. Appraisal of Evidence, policy evaluation, and trust decisions are performed by application-level components that consume the attestation properties exposed by the TLS stack. As a result, while no new application-layer protocol is required, applications do incorporate additional trust logic to interpret attested connection properties and make security-relevant decisions. Related to this, Markus Rudy shares his practical experience :
Conveying the evidence is not enough, it needs to be verified as well in order to end up with a trustworthy channel. We decided to integrate verification into the handshake, too, but that has massive drawbacks: Verification can take orders of magnitude longer than normal TLS handshakes, and usually involves remote calls, affecting all sorts of timeouts. However, doing the verification at the application level would require forwarding information from the handshake (e.g. nonce), at which point the application needs to be fully aware of the handshake protocol in order to verify it, breaking the intended layering.
Avoid Extra Round Trips for One-time Attestation It is claimed that intra-handshake attestation avoids extra round trips for use cases which require remote attestation only once during Connection Establishment Time. However, this may only be valid in cases when the Connection Establishment Time without remote attestation is significantly higher than the time for generation and appraisal of Evidence, such as cross-continent. For instance, Markus Rudy shares his practical experience :
I don't think saving extra roundtrips is an appropriate design goal when attestation is required. Generating evidence alone takes much longer than normal network roundtrip times, not even speaking of verification.
On request, he kindly conducted an experiment and shared his preliminary results of experiment based on attested TLS implementation in Edgeless Systems Contrast where Coordinator is one of the components : * SNP: Median time of 140ms from TCP SYN to TLS channel established and evidence sent to the client. > * TDX: Median time of 1020ms, same procedure. I don't know why it is that slow, it should only be making machine-local remote calls, if any. * So far, I only managed to measure TDX verification, which adds another 340ms. This is bound by remote HTTP requests, afaiu, and could be optimized with locally cached collateral, CRL, etc. I'd expect SNP to exhibit similar timing, because verification does similar remote calls. AMD is probably quicker because they're trading off with the appraisal time: the AMD report is not self-contained and can be generated with only the VM and the SP, but for verification you need to fetch the VCEK from somewhere, whereas the Intel quote includes the PCK cert and possibly other things that need to be fetched from a host service, if not the internet. ]]> We summarize that in the following table: Preliminary analysis by Markus Rudy (Median time in ms)
Property Intel TDX AMD SEV-SNP
Generation of Evidence + TLS 1020 140
Appraisal of Evidence 340 not available (expected ca. 340)
Note that since continuous attestation is generally required , early attestation adds unnecessary complexity. However, Yaroslav Rosomakho (as individual contributor) raises a concern :
The argument that avoiding an extra RTT is not a relevant goal may depend heavily on deployment topology (LAN vs same-region vs cross-continent).
Request for Contributions We invite the WG to submit their analysis results for cases such as:
  • within continent
  • across continent
Limitations
Limited Claims Availability Since limited Claims are available at the Evidence Generation Time, it does not provide complete security posture of the Attester, such as runtime integrity of Attester. Examples include dynamic Claims, such as weights of trained model and contextual data in the case of AI agents/agentic AI , , . These dynamic Claims are neither available for pre-handshake attestation nor for intra-handshake attestation.
Invasive Changes in TLS and Security Concerns To be made secure for confidential computing, it requires invasive changes in TLS protocol, as deep as key schedule and adding or modifying existing handshake messages , which are explicitly out of scope of :
The attested (D)TLS protocol extension will not modify the (D)TLS protocol itself. It may define (D)TLS extensions to support its goals but will not modify, add, or remove any existing protocol messages or modify the key schedule.
A detailed analysis of different binding mechanisms for intra-handshake attestation is available at and the corresponding formal analysis is available at .
State After Connection Establishment Not Covered It provides no guarantees about the state of Attester during the Lifetime of Connection. This is a security concern in long-lived connections where state of Attester (at workload or platform level) may change after Connection Establishment Time. Examples include AI agents/agentic AI , , . Note that session resumption is a new connection .
High Handshake Latency Because of signature in Evidence generation and verification of signatures during appraisal, this leads to high handshake latency. This may not be desirable for some applications. Markus Rudy shares his practical experience :
Conveying the evidence is not enough, it needs to be verified as well in order to end up with a trustworthy channel. We decided to integrate verification into the handshake, too, but that has massive drawbacks: Verification can take orders of magnitude longer than normal TLS handshakes, and usually involves remote calls, affecting all sorts of timeouts. However, doing the verification at the application level would require forwarding information from the handshake (e.g. nonce), at which point the application needs to be fully aware of the handshake protocol in order to verify it, breaking the intended layering.
As Yaroslav Rosomakho (as individual contributor) observes , note that Post-Quantum (PQ) transition may change the baseline. We argue that while PQ is unavoidable within TLS handshake, remote attestation is avoidable (see ).
Maturity of TEEs With several attacks (see ), attestation in TEEs may not yet be mature enough to be integrated within TLS handshake. Ayoub Benaissa remarks :
TLS might not be well suited to include this in its protocol. Not sure TEEs are even as mature for the people to see that it should be included right now. The plan to make it a post-handshake protocol makes more sense right now. A future where it's incorporated into TLS might exist, but I don't think there is enough motivation right now.
Amount of Effort Markus Rudy shares his practical experience :
"Keeping attestation out of the application logic" is not as straightforward as it sounds. In the background-check model, the attester needs to collect evidence in response to the relying party's challenge (nonce). We were lucky that the Golang TLS stack can be supplied with arbitrary closures that are called during the handshake, but in my experience this is a rare design choice and may also be difficult to implement in other languages.
Ayoub Benaissa remarks :
An intra-handshake requires much more work compared to a post-handshake. People need to agree on how to add this as optional in TLS (we can't force everyone to use it of course), the standard needs to be implemented by major libraries, and then it will be available in major client/server applications. If any of the prior steps doesn't go through, it means you have to patch your components to make it work, which is not convenient / less secure.
Difficulty of Debugging Attestation Markus Rudy shares his practical experience :
There's only so much information in a TLS alert message, and it's definitely not enough to understand remote verification failures. While I understand this to be a deliberate design choice by TLS, I found this to be a hindrance for operating and debugging a large number of services in practice.
Post-handshake Attestation Post-handshake attestation improves the situation further by signing the Claims during Lifetime of Connection, i.e., at the time when it is actually required. Hence, together with use cases requiring one-time attestation, it covers the use cases of long-lived connections requiring re-attestation. For post-handshake attestation, first round of remote attestation MUST be done immediately after Connection Establishment Time, and Relying Party (RP) MUST not send any secure data until Evidence is successfully appraised. As Yaroslav Rosomakho (as individual contributor) proposes :
an explicit shim or gating layer that performs attestation after the TLS handshake completes but before any application data is exchanged. This is operationally distinct from both intra-handshake attestation and fully application-integrated post-handshake attestation. Such an approach could preserve standard TLS handshake behaviour and latency characteristics, avoid invasive TLS changes, and still prevent application data from flowing until attestation succeeds. It may also mitigate some of the application-layer complexity by localizing attestation handling to a well-defined enforcement point (e.g., a sidecar or connection gate) rather than requiring per-protocol integration.
So a promising idea is to have an attested TLS library as a layer in between TLS implementation and application layer. Attested TLS library -> Application Layer ]]> In following subsections, we present the benefits and limitations of post-handshake attestation.
Benefits In general, it allows re-authentication and re-attestation without tearing down the connection.
Full Claims Availability Since all Claims are available at the time of post-handshake attestation (during Lifetime of Connection), it provides complete security posture of the Attester.
No Change in TLS It does not require any change in TLS protocol.
State After Connection Establishment Is Covered It provides guarantees about the state of Attester during the Lifetime of Connection. This is particularly helpful in long-lived connections where state of Attester may change after Connection Establishment Time.
Standard Handshake Latency Since the signature in Evidence generation and verification of signatures during appraisal happen after Connection Establishment Time, there is no additional latency. Yaroslav Rosomakho (as individual contributor) shares his concern :
I don't think that moving latency related to attestation into after handshake is always a good thing. In some real-time and streaming applications, a spike after the session is established may be much more disruptive than paying a cost during the handshake.
We believe this concern can be resolved by the layering described in .
Avoid Extra Round Trips Except for first round of remote attestation, post-handshake attestation outperforms the intra-handshake attestation (one round trip), which requires re-establishing the connection (1.5 round trip).
Ease of Implementation Ayoub Benaissa remarks :
We already implemented a post-handshake protocol and have a full demo working. We were able to do this in a matter of weeks. That's because you don't need to modify any TLS implementation, but only add a few verification steps after the usual TLS handshake. This is almost the same on the client and server side.
Production-grade deployments, including Google STET and SCONE , exist.
Ease of Verification and Audit Post-handshake attestation has relatively easier formal analysis and verification. The same may apply to audit. Markus Rudy remarks :
(Formal) verification of a protocol and audit of its implementations might be much easier if it ran on top of TLS. Existing proofs and certifications would not need to be reevaluated.
General Solution for Other Protocols In post-handshake attestation, design, verification and audit effort will be one-time and any protocol (e.g., Noise) which has support for exporters can then use it without changing each and every protocol. Markus Rudy shares this requirement :
It should be possible to port the general shape of a post-handshake attested TLS protocol to other protocols that provide secure channels and session binding (Noise comes to mind).
Limitations
Impact on Application Layer Post-handshake attestation may require changes at the application layer. However, changes at the application layer do not necessarily imply modifications to application business logic or data exchange protocols. Attestation-related functionality may be realized via application-level signalling (Exported Authenticators ) and trust logic, which may be implemented in intermediary components (e.g., proxies, sidecars, or middleware) on both client and server sides. These components are responsible for exchanging and appraising attestation evidence and enforcing trust or authorization decisions before application data is processed. This is analogous to common production deployments in which TLS termination and certificate handling are performed by a fronting proxy, while the application itself remains unchanged and resides behind it.
Need for Post-handshake Attestation We argue that post-handshake attestation is unavoidable (e.g., re-attestation to track changes after Connection Establishment Time for long-lived connections). Use cases where pre-handshake attestation and intra-handshake attestation are insufficient include AI agents/agentic AI . Intra-handshake attestation only adds unnecessary complexity which is avoidable. All identified use cases where intra-handshake attestation seems suitable can be covered by post-handshake attestation (by doing attestation round immediately after Connection Establishment Time) but not the other way around.
IoT Constraints includes TLS client as RATS Attester. Client could be a low-power IoT device. There are use cases where periodic or on-demand attestation is required, such as periodic attestation for long-lived, low-power IoT devices or in IoT swarms that need to synchronize software versions before coordinated operations or after configuration updates. Moreover, we note some observations from LAKE WG: Michael Richardson shares his insight :
I have a half-written document on putting EAT into the full BRSKI protocol. A reason that I stopped is that I realized that doing security posture evaluation at onboarding time (only) wasn't enough. It has to be done regularly. So having a protocol used at onboarding time and another one during normal operation meant that the onboarding one would have bugs that never get fixed, since the code only runs once.
He further shares :
My contention, which I think the group agreed with, is that one probably wants to do continuous assurance, that is, to repeat the remote attestation.
Do you want to have two protocols and two code paths? (redundant code in a constrained device?). I suggested that maybe the remote attestation should use it's own /.well-known Path, and that it would just occur after onboarding, and regularly onwards. Maybe it's weird to onboard a device only to kick it out again immediately because it failed remote attestation, but given continuous assurance, this could happen at any time.
Göran Selander observes :
Indeed, if the authentication procedure is repeated at a later stage, for whatever reason, e.g. key rotation, it should be possible to repeat the attestation procedure.
Existing Implementations
Intra-handshake Attestation Prominent implementations of intra-handshake attestation are all vulnerable to relay attacks . Some of them are abusing the extensions of TLS, such as SNI and ALPN, for conveyance of attestation nonce .
Post-handshake Attestation Google , Microsoft , and SCONE are all using post-handshake attestation.
Security Considerations Most of the document is about security considerations. Also, Security Considerations of and apply. In addition:
  • Pre-handshake attestation is vulnerable to replay and diversion attacks. See the corresponding formal analysis at . Moreover, pre-handshake attestation leads to a single point of failure.
  • Without significant changes to the TLS protocol: Intra-handshake attestation is vulnerable to diversion attacks . We reported these attacks to TLS WG in February 2025 . A formal proof is available for further research and development. Since reporting to TLS WG, the assumptions required for these attacks have been practically demonstrated in TEE.fail, Wiretap.fail, and BadRAM. More recently, we found that intra-handshake attestation also does not bind the Evidence to the application traffic secrets, resulting in relay attacks . A detailed formal analysis of binding mechanisms is available at . Analyzed implementations include Meta's AI, Cocos AI, Edgeless Systems Contrast, and CCC Attestation SIG's adopted project attested TLS proof of concept.
  • No attacks on post-handshake attestation are currently known. Post-handshake attestation avoids replay attacks by using fresh attestation nonce. Moreover, it avoids diversion and relay attacks by binding the Evidence to the underlying TLS connection, such as using Exported Keying Material (EKM) , as proposed in Section 9.2 of . and provide mechanisms for such bindings. Efforts for a formal proof of security of post-handshake attestation are ongoing.
Exploit of Sensitive Hardware-level Information From the view of the TLS server, post-handshake attestation offers better security than intra-handshake attestation when the server acts as the Attester. In intra-handshake attestation, due to the inherent asymmetry of the TLS protocol, a malicious TLS client could potentially retrieve sensitive hardware-level and firmware-level information from the server's Evidence without the client's trustworthiness (i.e., authentication) first being established by the server. This information (e.g., vulnerable firmware version) can be exploited for attacks on the server. In post-handshake attestation, the server can ask for client authentication during the handshake and only send the Evidence after successful client authentication. We believe this regression of intra-handshake attestation violates : Additionally, we are not convinced that these non-trivial changes in intra-handshake attestation will not break the well-established computational security properties of TLS in CryptoVerif (cf. this), as mentioned in Appendix F.1.6. of .
Detailed Security Analysis A thorough security analysis is presented in .
IANA Considerations This document has no IANA actions.
References Normative References Key words for use in RFCs to Indicate Requirement Levels In many standards track documents several words are used to signify the requirements in the specification. These words are often capitalized. This document defines these words as they should be interpreted in IETF documents. This document specifies an Internet Best Current Practices for the Internet Community, and requests discussion and suggestions for improvements. Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words RFC 2119 specifies common key words that may be used in protocol specifications. This document aims to reduce the ambiguity by clarifying that only UPPERCASE usage of the key words have the defined special meanings. Informative References Remote ATtestation procedureS (RATS) Architecture In network protocol exchanges, it is often useful for one end of a communication to know whether the other end is in an intended operating state. This document provides an architectural overview of the entities involved that make such tests possible through the process of generating, conveying, and evaluating evidentiary Claims. It provides a model that is neutral toward processor architectures, the content of Claims, and protocols. Exported Authenticators in TLS This document describes a mechanism that builds on Transport Layer Security (TLS) or Datagram Transport Layer Security (DTLS) and enables peers to provide proof of ownership of an identity, such as an X.509 certificate. This proof can be exported by one peer, transmitted out of band to the other peer, and verified by the receiving peer. Channel Bindings for TLS 1.3 This document defines a channel binding type, tls-exporter, that is compatible with TLS 1.3 in accordance with RFC 5056, "On the Use of Channel Bindings to Secure Channels". Furthermore, it updates the default channel binding to the new binding for versions of TLS greater than 1.2. This document updates RFCs 5801, 5802, 5929, and 7677. Perspicuity of Attestation Mechanisms in Confidential Computing: Technical Concepts The Transport Layer Security (TLS) Protocol Version 1.3 Independent This document specifies version 1.3 of the Transport Layer Security (TLS) protocol. TLS allows client/server applications to communicate over the Internet in a way that is designed to prevent eavesdropping, tampering, and message forgery. This document updates RFCs 5705, 6066, 7627, and 8422 and obsoletes RFCs 5077, 5246, 6961, 8422, and 8446. This document also specifies new requirements for TLS 1.2 implementations. Identity Crisis in Confidential Computing: Formal Analysis of Attested TLS TU Dresden, Dresden, Germany Aalto University, Espoo, Finland Aalto University, Espoo, Finland ACM Identity Crisis in Confidential Computing: Formal Analysis of Attested TLS Intra-handshake.fail (CVE-2026-33697): High-severity CVE in Attested TLS Intra-handshake.fail (CVE-2026-33697): High-severity CVE in Attested TLS Towards Validation of TLS 1.3 Formal Model and Vulnerabilities in Intel’s RA-TLS Protocol Faculty of Computer Science, Technical University of Dresden, Dresden, Germany Huawei Technologies Oy, Helsinki, Finland Department of Computer Science, University of Applied Sciences Bonn-Rhein-Sieg and Siemens, Sankt Augustin, Germany Linaro, Lausanne, Switzerland Institute of Electrical and Electronics Engineers (IEEE) Dynamic Attestation for AI Agent Communication This document describes a use case for conveying remote attestation information in association with Transport Layer Security (TLS) sessions in the context of AI agent communication. It focuses on long-lived secure channel sessions where an AI agent runtime posture, covering the platform Trusted Computing Base (TCB), agent manifest (models, tools and policies) and committed runtime context, can change frequently and unpredictably. The document highlights requirements for dynamic attestation so that relying parties can base authorization decisions on the current runtime posture of the communicating agent. Secure Evidence and Attestation Transport (SEAT): Charter for Working Group Re: Comments for remote attestation over EDHOC Evaluation of attestation results by EDHOC clients Re: Comments for remote attestation over EDHOC Split-Trust Encryption Tool Attested Session Handling Azure vTPM Attestation and Binding SoK: Attestation in Confidential Computing Re: New Version Notification for draft-usama-seat-intra-vs-post-00.txt Re: New Version Notification for draft-usama-seat-intra-vs-post-00.txt Impersonation attacks on protocol in draft-fossati-tls-attestation (Identity crisis in Attested TLS) for Confidential Computing Identity Crisis in Confidential Computing: Formal analysis of attested TLS protocols Security Goals and Use Cases for Integrating Remote Attestation with Secure Channel Protocols Arm TU Dresden Linaro Nokia China Mobile This document outlines desirable security goals and use cases for integrating remote attestation (RA) capabilities with secure channel establishment protocols (e.g., TLS and DTLS). Peer authentication in such protocols establishes trust in a peer's network identifiers but provides no assurance regarding the integrity of its underlying software and hardware stack. Remote attestation addresses this gap by enabling a peer to provide verifiable evidence about the current state of the Target Environment. This document specifies a set of essential security goals the protocol solution must have, including cryptographic binding to the secure connection, evidence freshness, and flexibility to support different attestation models. It then explores relevant use cases, such as confidential data collaboration and secure secrets provisioning, to motivate the need for this integration. This document is intended to serve as an input to the design of protocol solutions within the SEAT working group. Re: New Version Notification for draft-usama-seat-intra-vs-post-00.txt Re: Requesting review of IETF draft on categories for attested TLS Re: New Version Notification for draft-usama-seat-intra-vs-post-02.txt [Use Case] RATS for Hardware-Enforced State Management in Autonomous Agents (AIGA) AI Governance and Accountability Protocol (AIGA) This document specifies the AI Governance and Accountability (AIGA) Protocol version 1.0, a practical, economically viable, and technically enforceable framework for governing autonomous AI agents. AIGA 1.0 is designed to address real-world deployment constraints, adversarial agent scenarios, and economic incentive alignment. The protocol is founded on a Tiered Risk-Based Governance model, applying proportional oversight to agents based on their capabilities. All agents are governed by an Immutable Kernel Architecture which provides a non-modifiable Trusted Computing Base (TCB) for enforcing policy. This is combined with Action-Based Authorization, where critical operations require real-time approval. To solve the single-point-of-failure problem, the protocol uses a Federated Authority Network of regional, cross-validating hubs and provides a Network-Level Quarantine Protocol for enforcement. The entire framework is designed around Economic Incentive Alignment, making compliance the most economically rational choice for operators. For high-assurance (T3-T4) scenarios, AIGA 1.0 specifies advanced, redundant mechanisms including Multi-Vendor TEE Attestation (M-TACE), AI "Warden Triumvirate" Triage, Human Review Board (HRB) Multi-Signature, Peer Consensus Failsafe & Identity Rotation, and Double Ratchet Cryptography. JSON Web Token (JWT) JSON Web Token (JWT) is a compact, URL-safe means of representing claims to be transferred between two parties. The claims in a JWT are encoded as a JSON object that is used as the payload of a JSON Web Signature (JWS) structure or as the plaintext of a JSON Web Encryption (JWE) structure, enabling the claims to be digitally signed or integrity protected with a Message Authentication Code (MAC) and/or encrypted. HTTP Message Signatures This document describes a mechanism for creating, encoding, and verifying digital signatures or message authentication codes over components of an HTTP message. This mechanism supports use cases where the full HTTP message may not be known to the signer and where the message may be transformed (e.g., by intermediaries) before reaching the verifier. This document also describes a means for requesting that a signature be applied to a subsequent HTTP message in an ongoing HTTP exchange. Assertions and Protocols for the OASIS Security Assertion Markup Language (SAML) V2.0 MITRE's New Framework: Securing the eBPF Layer Your AI Depends On Framework for Continuous Remote Attestation Early Attestation Considered Very Harmful (CVE-2026-100833, CVE-2026-100835,CVE-2026-108265, CVE-2026-108266, CVE-2026-108267, CVE-2026-108268, CVE-2026-108269, CVE-2026-92701, CVE-2026-92702, CVE-2026-108362, CVE-2026-33697, CVE-2026-108361 and others) Global Alliance for Genomics and Health CoreWeave University of Namur Stevens Institute of Technology Independent Shanghai Guan An Information Technology Co., Ltd. Switch DPhil Oxford University Sureshot Labs Independent Independent Corporate Risk Consultant PHI-OMEGA HashCloak Inc and Stoffel Labs Inc EMILIA Protocol, Inc. Ecole Superieure Ingénieurs Léonard de Vinci Paris Birmingham City University Verily The draft aims to provide technical details of [CVE-2026-100833] of CVSS 9.6, [CVE-2026-108265] of CVSS 9.1, [CVE-2026-108266] of CVSS 9.1, [CVE-2026-108267] of CVSS 9.1, [CVE-2026-108268] of CVSS 9.1, [CVE-2026-108269] of CVSS 9.1, [CVE-2026-100835] of CVSS 9.1, [CVE-2026-92701] of CVSS 9.1, [CVE-2026-92702] of CVSS 9.1, [CVE-2026-108362] of CVSS 9.1, [CVE-2026-119344] of CVSS 9.1, [CVE-2026-33697] of CVSS 7.5, [CVE-2026-108361] of CVSS 7.2, [CVE-2026-119345] of CVSS 7.2, and 26+ GitHub Security Advisories (GHSAs) against early attestation in the process of getting CVE IDs, which provide substantial technical evidence of how early attestation, such as [I-D.fossati-seat-early-attestation], fails in practice, even *without physical access* to the desired machine. Moreover, since continuous attestation is generally required [CSA-eBPF] [MITRE-Continuous-Attestation], early attestation adds *unnecessary complexity*. The results are backed by the research [Intra-handshake.fail], [TLS-RA], [EarlyAttestationBleed], [EarlyAttestationTschuess] and the artifacts [Intra-handshake.fail-repo] in state-of-the-art formal analysis tool, ProVerif, under Apache-2.0 license for reproducibility, extensibility, and review, and have been acknowledged by the relevant stakeholders. Currently, there is *one CVE of CVSS 9.6, 11 CVEs of CVSS 9.1, one CVE of CVSS 8.2, one CVE of CVSS 7.5, one CVE of CVSS 7.2, and 26+ GHSAs published awaiting CVE IDs against the broader early attestation covering all layers of the ecosystem up to the application*. The research papers on these are currently either under submission or being prepared for submission. The artifacts of these papers will be shared with the community under Apache-2.0 license for reproducibility, extensibility, and review. Based on our work, all except two implementations of early attestation have been archived, withdrawn, or moved to post-handshake attestation. In our analysis [Intra-handshake.fail-repo], [ID-Crisis-repo], and [EarlyAttestationTschuess] (EAT) and based on discussions in the CCC Attestation SIG, IETF SEAT WG, IRTF UFMRG, and several other forums, the remaining two implementations of early attestation -- Edgeless Systems Contrast and Meta's AI -- remain vulnerable due to lack to binding to the underlying connection. We recommend users of these two implementations to carefully evaluate their systems and understand the risks. If you are aware of any open-source implementation of early attestation that is not yet covered in this draft, please let us know to responsibly disclose the vulnerabilities to them.
Acknowledgments We gratefully thank the following:
  • Peg Jones for review of early draft before submission of -00
  • Paul Wouters for review of section 4 of -00
  • Ayoub Benaissa for review of -00 and sharing his practical experiences
  • Markus Rudy for review of -00 and sharing his practical experiences and for conducting experiments with TDX and SNP on our request
  • Mike Bursell (Executive Director, Confidential Computing Consortium) for review of -01
  • Yaroslav Rosomakho (as individual contributor) for detailed review of -02
Contributors Pavel Nikonorov (GENXT / IIAP NAS RA) contributed text in and .
History -01
  • Added scope section to address comments of Paul Wouters
  • Added comments of Ayoub Benaissa as quotes
  • Added some subsections to incorporate practical experiences of Markus Rudy
  • Extended security considerations
-02
  • Added experiments by Markus Rudy
  • Removed our experiments
  • Added reference to use cases document to address comment of Mike Bursell
-03
  • Added advantages of pre-handshake attestation
  • Added references for limited Claims for intra-handshake attestation
  • Added Yaroslav's proposal in post-handshake attestation