| Internet-Draft | RATS Security Considerations | November 2026 |
| Sardar, et al. | Expires 6 May 2027 | [Page] |
This document aims to provide guidelines and best practices for writing security considerations for technical specifications for RATS targeting the needs of implementers, researchers, and protocol designers. In particular, it discusses some of the 'bottom turtle' issues. This is a work-in-progress, and the current version mainly presents an outline of the general security guidelines, baseline, or template for RATS that future versions will cover in more detail.¶
This note is to be removed before publishing as an RFC.¶
The latest revision of this draft can be found at https://muhammad-usama-sardar.github.io/rats-sec-cons/draft-sardar-rats-sec-cons.html. Status information for this document may be found at https://datatracker.ietf.org/doc/draft-sardar-rats-sec-cons/.¶
Source for this draft and an issue tracker can be found at https://github.com/muhammad-usama-sardar/rats-sec-cons.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 6 May 2027.¶
Copyright (c) 2026 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
Every Internet Draft needs to have a "Security Considerations" section. While general guidelines such as [RFC3552] exist, the underlying threat model is that the endpoint is fully trusted (i.e., all software and hardware components in the device may access the keys). RATS [RFC9334] has a primarily different threat model in the sense that only parts of the endpoint (called Attester) are trusted (i.e., only specific software and hardware components in the device may access the keys), and the goal is to establish the trustworthiness of the endpoint. In other words, [RFC3552] deals with a network adversary, whereas RATS deals with an endpoint adversary, which may have root access or physical control over the device with which it can extract keys from software or hardware.¶
Moreover, remote attestation has several distinguishing features that necessitate a separate document. One specific example of such a feature is the architectural complexity of the endpoint. While network protocols typically have 2 roles, RATS has additional roles, which complicates the picture. Unfortunately, no guidelines currently exist for remote attestation [RFC9334] in RATS. This document aims to fill this gap.¶
Moreover, while the target audience of Internet Drafts is implementers, researchers, and protocol designers [I-D.irtf-cfrg-cryptography-specification], RATS drafts generally do not fulfill these needs, in particular the needs of researchers and protocol designers. On the other hand, in our observation, implementers generally find it hard to relate the abstract concepts of RATS to the real-world systems. In general, implementers and protocol designers of RATS are thus left with little or no guidance.¶
Unverified protocol designs, imprecisely stated threat model and security goals have led to high and critical severity vulnerabilities related to remote attestation.¶
The formal analysis led to three orthogonal issues:¶
Formal analysis [ID-Crisis-repo] found diversion attacks when unique hardware identifier is not included in Evidence. For technical details, please see the corresponding paper [ID-Crisis].¶
Formal analysis [Intra-handshake.fail-repo] of several production implementations of remote attestation led to the discovery of [CVE-2026-33697] of CVSS 7.5 for relay attacks. For technical details, please see the corresponding paper [Intra-handshake.fail].¶
Further formal analysis of production implementation of remote attestation has led to discovery of another class of attacks and will potentially lead to three CVEs (currently under responsible disclosure) each with an expected CVSS 9.1.¶
This shows the value of precise threat model and formal analysis in the design of secure protocols to find subtle vulnerabilities, which could otherwise be missed. This draft aims to provide the baseline security considerations that other drafts can simply refer to.¶
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 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶
Authentication is a term which is often ambiguous in RATS specifications. We propose general hierarchy of one-way authentication, which can help precisely state the intended level of authentication (in decreasing order):¶
Recentness can be added to each of these levels of authentication. Details will be added in future versions.¶
This section describes "What can go wrong?"¶
See Section 4 of [Intra-handshake.fail] as an example.¶
It has both legal and technical perspective.¶
Data subject is an identifiable natural person (as defined in Article 4 (1) of GDPR [GDPR]).¶
(Data) Controller (as defined in Article 4 (7) of GDPR [GDPR]) manages and controls what happens with personal data of data subject.¶
(Data) Processor (as defined in Article 4 (8) of GDPR [GDPR]) performs data processing on behalf of the data controller.¶
Infrastucture Provider is a role which refers to the Processor in GDPR. An example of this role is a cloud service provider (CSP).¶
See Section 6.1 of [Intra-handshake.fail] as an example.¶
Security considerations in RATS specifications need to clarify how the following attacks are avoided or mitigated:¶
In this attack, a network or endpoint adversary -- with access to older Evidence -- can replay Evidence with stale Claims which no longer represent the actual state of the Attester, potentially resulting in exposure of confidential data [RA-TLS].¶
Replay of stale Evidence may be within the same connection or across multiple connections.¶
In this attack, a network adversary -- with Dolev-Yao capabilities [Dolev-Yao] and access (e.g., via Foreshadow [Foreshadow]) to the attestation key of any machine in the world -- can redirect a connection intended for a specific Infrastructure Provider to the compromised machine, potentially resulting in exposure of confidential data [ID-Crisis].¶
In the context of confidential computing and TLS as a transport protocol, we reported these attacks to the TLS WG in February 2025 [Usama-TLS-26Feb25]. A formal proof is available [ID-Crisis-repo] for further research and development. Since reporting to TLS WG, these attacks have been practically exploited in TEE.fail, Wiretap.fail, and BadRAM.¶
In this attack, a network or endpoint adversary -- with access to suitable binding material -- can relay an attestation request to a genuine Attester and present the genuine Evidence as its own, potentially resulting in impersonation of genuine Attester [Intra-handshake.fail].¶
Note that replay is about same Attester while relay attack is about different Attesters.¶
[Intra-handshake.fail] presents relay attacks in early attestation. From a security perspective, one of the key decision factors is the candidate binding mechanism. The formal analysis artifacts are available in [Intra-handshake.fail-repo] under Apache-2.0 license for reproducibility, extensibility, and further research.¶
This work resulted in [CVE-2026-33697] of CVSS 7.5.¶
A complementary paper [ID-Crisis] presents the identity crisis in pre- and intra-handshake (aka early) attestation. The formal analysis is available in [ID-Crisis-repo] under Apache-2.0 license for reproducibility, extensibility, and further research.¶
[EarlyAttestationBleed] presents a formal analysis together with regression tests of the broader attestation ecosystem and discovered three critical-severity vulnerabilities in implementations of early attestation:¶
Ultraviolet Cocos AI in TDX path resulting in [CVE-2026-92701] of CVSS 9.1.¶
Ultraviolet Cocos AI in AMD SEV-SNP path resulting in [CVE-2026-92702] of CVSS 9.1.¶
Edgeless Systems Contrast in policies resulting in [GHSA-Edgeless-Systems2] of CVSS 9.0-10.0 and [CVE-2026-100833] of CVSS v3.1 of 9.6.¶
Sec. 7 of [EarlyAttestationBleed] presents four architectural flaws in early attestation:¶
A malicious TLS client retrieves sensitive hardware-level and firmware-level information from the server’s Evidence before the server has authenticated the client.¶
Early attestation can typically provide information about only boot-time Claims, since most runtime Claims may not be available at the handshake time.¶
Re-attestation requires post-handshake attestation. Early attestation thus adds unnecessary complexity.¶
Early attestation results in high handshake latency, timeouts, and even failures.¶
[EarlyAttestationTschuess] resulted in the following CVEs demonstrating relay and diversion attacks against early attestation implementations/deployments:¶
[CVE-2026-100835] of CVSS 9.1 against Edgeless Systems Contrast¶
[CVE-2026-112377] of CVSS 9.1 against CCC Attestation SIG¶
[CVE-2026-108362] of CVSS 9.1 against Privasys¶
[CVE-2026-108265] of CVSS 9.1 against Privasys¶
[CVE-2026-108266] of CVSS 9.1 against Privasys¶
[CVE-2026-108267] of CVSS 9.1 against Privasys¶
[CVE-2026-108268] of CVSS 9.1 against Privasys¶
[CVE-2026-108269] of CVSS 9.1 against Privasys¶
Further examples of CVEs are in [I-D.intra-handshake-fail].¶
This section describes the countermeasures and their evaluation.¶
To mitigate the above attacks, we propose post-handshake attestation. Post-handshake attestation avoids replay attacks by using a fresh attestation nonce. Moreover, considering TLS as the transport protocol, it avoids diversion and relay attacks by binding the Evidence to the underlying TLS connection, such as using Exported Keying Material (EKM) [I-D.ietf-tls-rfc8446bis], as proposed in Section 9.2 of [ID-Crisis]. [RFC9261] and [RFC9266] provide mechanisms for such bindings. Efforts for a formal proof of security of post-handshake attestation are ongoing.¶
All of this document is about security considerations.¶
This document has no IANA actions.¶
The authors wish to thank Ira McDonald and Ivan Gudymenko for insightful discussions.¶