[Seat] Re: Comments on formal analysis of relay attacks in attested TLS (CVE-2026-3369)

[Markus Rudy <[e-mail]>]

Hi Usama,

> I think we are largely talking past each other.

I agree - let's try to understand where and why.

> 1. Unless I am misunderstanding something, that is not really "missing" in the paper since that is the proposed binder in Sec. 7.2 of [0], and the results are already in Table 4 of [0]. ProVerif artifacts are in folder 'proposal' of [3].

Thanks, I don't know how I managed to miss that, sorry. The paper states:

"We prove in
ProVerif that it achieves level 2 (G2), as shown in Table 4, whereas G3 evaluates
to false. Our observation from the TLS key schedule is that at the point in
time when the intra-HS binder is created, kc is not yet derivable. Based on this
observation and our extensive discussions with the TLS WG, we believe that it
may not be possible to achieve level 3 in intra-HS attestation while preserving
the established security and privacy properties of the TLS protocol."

If G3 evaluates to false, you should be able to provide a counter-example that shows how an adversary can come in possession of the traffic secrets while the session is bound to the handshake secret. I'd be interested how that vector looks like, because it would contradict my proof directly.

> 2. I believe I covered a good number of intuitive arguments in slide 2 of my SEAT presentation [1]. 

I think what's missing in that slide are two observations: 

- If the handshake secret is known to the attacker, that attacker can compromise the application traffic secrets. So the handshake secret security is not "irrelevant for security goals".
- The server may not be authenticated at the point where evidence is generated, but it is authenticated after the TLS handshake completes. That's my argument: if you look at the entire state of the TLS session, after it's created, it's enough to observe binding to the handshake secret and authenticating the server as in standard TLS.

> 3. I'm not sure what question you are trying to settle for SEAT, where the charter says [...]

This discussion came up several times on the list, and I don't think "deriving a binder from a TLS key" is considered an extension of the key schedule by everyone. Is EKM an extension of the key schedule, too? Deriving another key from the exported key material?

> Thanks for the clarification. I believe Table 4 in [0] has clear counter-examples to your argument. For instance, there are binders which satisfy G1 but not G2 and G3.

I'm sorry to say, but that is a table with emojis, not a clear counter-example. You may be aware of the concrete counter examples, but they are not very accessible in their current form. It would be enlightening to see the actual counter-examples, because then we could understand whether we're missing considerations from standard TLS security in the formal analysis.

> Just to make sure you are looking at the right figure, I mentioned Fig. 3 of [0] which is the protocol and not the TLS key schedule. So I am not sure why you are mentioning "not part of the TLS key schedule."

Sorry for the imprecision, but I was hoping the rest of my mail somehow conveyed the message: this is not specific to TLS! Nothing changes in that picture! Assurance of non-LEK is guaranteed out of band!

I'm going to answer the following questions from my product's POV, but there may be other interpretations that make sense.

> 1. What exactly is the server identity in your view?

The hardware identity (Platform instance identity for TDX). This is unique, bound to a specific server and can't be forced by an attacker on different hardware.

> 2. Who assigns this identity?

Intel.

> 3. How is that identity supplying entity trusted?

I'm going to interpret this as "how is the PIID known to the verifier", since the ID supplier is trusted anyway in this case. Two ways:

- I'm running my own datacenter, and when I set up a new server I record the PIID into my verifier database.
- I'm running on hardware provided by my CSP. The CSP can simply publish a list of known PIIDs; or they can cross-sign PCK certificates to endorse the machines they own and operate (this is what POE does).

> 4. Where in the Evidence is the server identity conveyed? (exact field in Quote and Report of Intel TDX and AMD SEV-SNP)

This is part of the AK certificates, i.e. the PCK or the VCEK. Which is why I keep saying that the issue is in remote attestation per se, and can be mitigated in the verifier. It does not need to be considered in the TLS integration.

> I am not sure what you are talking about, and how this is related to the paper and this discussion. We are not comparing with and without TEE. In the world I live in, security is always evaluated compared to the claimed security properties. Confidential computing made a claim that there is no need to trust the cloud provider, and we are saying this is not possible in the current technologies.

I don't think we need to discuss the discrepancy between marketing and reality here - we agree on that. However, if we take "Cloud provider does not need to be trusted to some extent" as a mandatory security property, we will not be able to produce any secure protocol. This is very intuitive to understand: the CSP can mount a hardware attack, which is out of scope for TEE security properties, and either impersonate a TEE or extract secrets (not only EK, but also the traffic secrets). What I'm saying is that trusting the CSP does not defeat the purpose of TEEs (at least not entirely). 

Cheers, Markus







________________________________________
From: Muhammad Usama Sardar
Sent: Monday, July 27, 2026 1:30 AM
To: Markus Rudy; [e-mail]
Cc: [e-mail]
Subject: Re: [Seat] Comments on formal analysis of relay attacks in attested TLS (CVE-2026-3369)


Hi Markus,
I think we are largely talking past each other.
Let me summarize what you seem to be saying: you seem to believe that there are binary levels: level0 (no binding) and level1 (G1 <=> G2 <=> G3). Is that correct?
On the contrary, we show in paper that there four fine-grained levels: level0 (no binding); level1 (G1); level2 (G2) and level3 (G3). Results in Table 4 of [0] provide concrete counter-examples to your levels. We'll happily clarify this in the extended technical report with intuition and examples.
On 26.07.26 21:42, Markus Rudy wrote:
If someone thinks a specific binding mechanism is missing that might lead to different results for intra-handshake attestation, please let us know, and we will happily share the analysis with the WG.

The binding mechanism I have in mind is the handshake secret, or something derived from it. Would be nice to see how this breaks down under formal analysis - I've not seen an argument that would explain that intuitively.
Thanks, that is helpful. A few notes:
Unless I am misunderstanding something, that is not really "missing" in the paper since that is the proposed binder in Sec. 7.2 of [0], and the results are already in Table 4 of [0]. ProVerif artifacts are in folder 'proposal' of [3].
I believe I covered a good number of intuitive arguments in slide 2 of my SEAT presentation [1]. For example, I removed the whole encryption done by handshake traffic key, and nothing changes in the security properties of TLS. Doing the same with encryption done by application traffic key literally beats the whole purpose of TLS; like why do TLS at all if all you want to do is to send application data unencrypted. I don't know how else to explain it more intuitively. Maybe someone else can phrase it better than me.
I'm not sure what question you are trying to settle for SEAT, where the charter says (emphasis my own):
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.
    Deriving keys from handshake secret is an extension of the key schedule which would already violate the SEAT charter.
Please see Sec. 6.4 in paper [0] which explicitly proves it cryptographically, and let us know what specifically you disagree with. Just vaguely disagreeing is not very helpful.

I thought I was specific: you proved G3 => G2 => G1, but you don't prove it strictly, such that G1 =!> G3. I'm arguing that G1 <=> G2 <=> G3, which does not contradict your prove, but makes it more precise.
Thanks for the clarification. I believe Table 4 in [0] has clear counter-examples to your argument. For instance, there are binders which satisfy G1 but not G2 and G3.
In the extended technical report, we will add more detailed traces and intuitive explanations for each binder, and disprove G1 => G2 => G3.
Please see first paragraph of G_3 in Sec. 6.3 in paper [0], which explains explicitly why it is an essential goal.

What I see in the paper is the following paragraph:

"After establishing an attested TLS connection, the client’s secrets (such as weights and context
data for inference) are encrypted using a key derived from the client’s application
traffic key atsc. Hence, an essential security goal is to analyze the correlation
of Evidence with atsc. The derivation of atsc includes handshake messages up
to the server’s F IN. Hence, G3 is at least as strong as G2."

That paragraph is entirely correct! However, it _does not rule out_ that you can _correlate_ to atsc by _binding_ to htsc. This is the same argument as above: (G1 <=> G3) would imply you can bind to any and correlate to atsc. Note that you are using "at least as strong" exactly as I mean it: it is as strong, but might not be strictly stronger.
As a concrete counter-example, there exists a binder (namely the one proposed in Sec. 7.2 of [0]) which satisfies G1 and G2 but not G3. I believe I covered a good number of arguments in slide 2 of my SEAT presentation [1].
Please see #4 in Sec. 7.1 of [0] for several practical reasons. Do we want that LEK in an unrelated server anywhere in the world breaks our connection? That is clearly too bad, and the world is probably better with standard TLS than have such a broken design of attested TLS.

I'm aware that the relay attack for the binding mechanisms you analyzed applies to any unrelated server being compromised. The mistake is the assumption that servers are cattle, and that you can't tell between a server compromised in someones basement and a server at your contractual $CSP. That, however, is not true: I can configure my evidence verification to only consider trusted (known) server hardware in the first place. This is independent of attested TLS, but a function of the verifier.
I disagree. Developers who are used to TLS must be explicitly given this guidance. Please see the chartering time discussion where IESG explicitly requested operational considerations.
Even if you consider configuration outside the scope of attested TLS, having configuration is insufficient. Somehow the identity of server hardware must be sent during the protocol to match against the configured trusted (known) server hardware.
We would be happy if you can share precisely what changes in Fig. 3 of [0] in the mitigations you have applied.

As I tried to explain several times, including in the post you responded to: this defect is _not part of the TLS key schedule_ - it's a shortcoming of remote attestation with current generation of TEEs! It arises due to the hardware vendors' threat model which does not cover everything that CC once advertised for. The mitigation removes the LEK attack vector, which the relay attack relies on. 
Just to make sure you are looking at the right figure, I mentioned Fig. 3 of [0] which is the protocol and not the TLS key schedule. So I am not sure why you are mentioning "not part of the TLS key schedule."
Also, as I mentioned, the statements/presentations/answers of Intel and what is written in their specifications are all very contradictory.
Continuing the idea of the last response above, I would like to see precise answers without handwaving to:
What exactly is the server identity in your view?
Who assigns this identity?
How is that identity supplying entity trusted?
Where in the Evidence is the server identity conveyed? (exact field in Quote and Report of Intel TDX and AMD SEV-SNP)
Second, this requires trusting the cloud provider, contrary to the whole claim of confidential computing.

We agree on that in principle, but not everything is so black and white. Running in a TEE still reduces the attack surface by a lot. Physical attacks in a hyperscaler datacenter are much harder to perform than software attacks (bringing a suitcase onto the floor and hooking up a machine vs. SSHing into it remotely). TEEs still protect from co-tenants that managed to breach their containment and gain software root on the hypervisor.
I am not sure what you are talking about, and how this is related to the paper and this discussion. We are not comparing with and without TEE. In the world I live in, security is always evaluated compared to the claimed security properties. Confidential computing made a claim that there is no need to trust the cloud provider, and we are saying this is not possible in the current technologies.
Third, what we report is the binding weakness. We do not believe it can be reasonably eliminated by anything other than changes in the protocol.

My entire message was about that. I do think there is an option for binding securely - the handshake secret - and I made both intuitive and mathematical arguments for why that holds. These should be countered with examples, rather than beliefs.
Please see my three points in the beginning and please answer them individually as precisely as possible.
If these CVEs are related to the discussion at hand I'd like to remind you: Edgeless uses a scheme you publicly call broken (correct), and I explained the mitigation we're using and why I think this invalidates the claim, looking at the whole picture. If you think this mitigation is not enough, you are cordially invited to responsibly disclose the reason to me, too.
Will follow-up off-list
Best regards,
Usama, Slava, and Jean-Marie
[0] https://www.researchgate.net/publication/408219182_Intra-handshakefail_CVE-2026-33697_High-severity_CVE_in_Attested_TLS
[1]: https://datatracker.ietf.org/meeting/126/materials/slides-126-seat-binding-properties-of-expat-00

[2] https://www.researchgate.net/publication/398839141_Identity_Crisis_in_Confidential_Computing_Formal_Analysis_of_Attested_TLS
[3] https://github.com/muhammad-usama-sardar/intra-handshake.fail