Hacker News new | ask | show | jobs
The RCE that AMD wouldn't fix (mrbruh.com)
215 points by MrBruh 8 hours ago
See also https://www.youtube.com/watch?v=4HjWHNLRMB0

Related: The RCE that AMD won't fix - https://news.ycombinator.com/item?id=46906947 - Feb 2026 (173 comments)

25 comments

> Final update: A couple of days before the embargo ended (and after I wrote the majority of this blog post), AMD told me what their patch for this vulnerability is [...] Although it is true that they now fully use HTTPS, the claim about signature verification is untrue; they only perform a CRC-32 check on the downloaded executable, which is not cryptographically secure.

So solves the MITM, but massive infection is still trivial if someone compromises the webserver.

> someone compromises the webserver

Sure, but that's true for 99% of things. Unless you establish trust outside of the normal distribution channel how would you protect against this? What is your proposed channel that is not bootstrapped from HTTPS PKI?

AMD (and Intel and everyone else) processors already have an HSM inside for confidential computing so use that? I would hope the HSM isn't as badly implemented as this update mechanism, but then again ...
AMD Software Engineers giving AMD Stupid Gaming Accessory Software Engineers access to a signing system backed by PSP seems like a much worse outcome than trusting HTTPS, really. Like, there are definitely intelligent and secure ways to do this, but this one in particular is overkill with a huge blast radius when it is (invariably) done incorrectly.
Those have been broken again and again. Even if not, how do you distribute the public keys for it, how do you bootstrap that trust?
Confidential computing is a whole thing with a key in each processor and a chain of trust and a way to remotely attest that your software is running in a secure enclave. All the vendors do it differently (sadly) but it's very much a solved problem.
There was a time when RDRAND on Zen gave all zeroes, or something, so eh...

I'm happy enough with TLS introduced: knowing the server I'm reaching for updates is actually 'amd.com'. Signatures would be nice, sure, but I wouldn't consider them nearly as critical or useful until now. Before we get too caught up in signatures, however, I'd like to see their new/improved updater actually take precedence.

As things stand, I'm not sure key rotation would go well... the updater doesn't mind itself, apparently.

What? The bootstrapping happened already! The official and correct AMD software already running on the computers. Preventing a human from falling for an impostor-website with malicious Download Now links is a separate problem.

The basics are straightforward: It'd be better if the current installation contains one (or more) public keys, and anything it downloads must validate as being signed by a corresponding private key. You don't need to do fancy things like global certs, discoverable keys, or revocation lists.

If today's installation doesn't have those checks and relies solely on HTTPS... well, that's unfortunate, but it's not like it poses some tricky dilemma! You simply use today's not-so-secure mechanism to get the new code which improves things going forward.

> The current installation shall already contain one (or more) public keys that it trusts for updates

The current installation was fetched via HTTPS, right? Either by you or in the factory.

Just saying the "bootstrapping already happened" does not make it not happen. It still needs to bootstrap trust from somewhere

I still can't figure out what problem you believe needs-fixing or what process you think needs to be designed. My most-charitable guesses are:

A. You're asking what should be done if the manufacturer's auto-update server has already been completely compromised by hackers and remains compromised.

B. [Implicitly rejected in last coment] You're asking how anybody can guarantee the very first install can be trusted even if someone has compromised drivers.amd.com .

C. You're asking if the auto-update process can somehow trick a compromised daemon into overwriting itself with a legit copy.

Those are all interesting to contemplate, but they are at best "out of scope".

[Followup] To over-communicate in the hope that it somehow resolves things, we already have this chain of trust:

1. Axiom: We trust the current daemon and OS. We must assume this, because otherwise it's an entirely separate problem and this whole discussion of an auto-update channel is irrelevant.

2. Axiom: We trust the owner. Tampering with the local auto-update process is not part of our threat-model, because a user who can do that doesn't need to.

3. The daemon is already coded to trust a replacement/successor installer if it meets certain criteria, which are:

3a. It comes from a trusted domain name it already knows should be owned by the same developer/company.

3b. The remote end is authenticated to "be" that domain via certificates from the (trusted) OS.

3c. The content is protected from tampering due, becauese we trust that TLS/SSL encrypts it.

That all already exists, it does not need to be torn down or rebooted. The proposal here is to simply to harden it with a new requirement in the next version:

3d. The next install must be signed by a trusted key-pair that was shipped with the current install.

This improves trust because it means an attacker would also need to compromise keys held in a release pipeline, which is much easier to secure than a CDN/webserver.

Sure, but the OEM is the definition of a ‘trusted environment’. They literally are assembling the equipment, if you can’t trust that, nothing else can be trusted from that point on anyway.
> someone
"Things break, Colonel!"
what are the chances of them caring so little, but implementing a dedicated signing server, HSM,etc..? even if they sign it, it will probably be done on the same web server.
It's ridiculous to consider MITM attacks out of scope for taking over your computer. Also, there are probably ways to exploit this without a true MITM like DNS cache poisoning. But it's best to just assume the whole internet is MITMed.
It's not out of scope "for taking over your computer". It's out of scope for the specific goals of the bug bounty program. Bug bounties are (usually) about prioritizing internal engineering effort; they are to vulnerability remediation what market feedback is to feature/function decisions in the rest of the product.

Everyone's judging this by the standard of "how good a bug" this is. But that's not necessarily how a bug bounty should function. Important prior to frame this with: neither any individual bug bounty submission nor the sum of all valid submissions materially alters the security of a serious product, at least not on their own. The system they feed into (for instance: security engineers taking a validated bounty submission and then quickly auditing the entire tree for variants of the same bug) can move the dials. The bounty bugs themselves though are mostly a sideshow.

What's especially weird (you didn't say this, but the sentiment has popped up on all 3 threads about this story) is the idea that AMD would be trying to cover this up. Why would they care? They run a bug bounty program. They've accepted the premise that they have vulnerabilities.

(From earlier today, in add'n: https://news.ycombinator.com/item?id=48492908).

But it should be their job to protect against MitM in their threat model. There is no rational reason to exclude them from the bug bounty. Doing so only leaves MitM attacks like this undisclosed.
I just gave a rational reason to exclude them from the bug bounty, which I can summarize as "the bug bounty is not their entire security program and does not have the goal you've axiomatically derived for it".

Cards on the table I am not a fan of bug bounty programs, and the fact that they're an engineering process that turns out to be impossible to have public engineering discussions about is definitely one of many reasons why. Most companies should not run bug bounty programs.

MITM where attacker needs to install their own CA certs on the victim's device -- sure, out of scope.

MITM because you used http instead of https and you don't have any other verified cryptographic signature on your data -- get tae fuck, fix it pronto.

I'd even count this as "having local access to the device", as that is what is needed to install such a cert
I think it's fair to say that requiring local administrative access to the device is out of scope, since you have already completely pwned the device in that case, which is what what you need to install a CA cert on any OSes.
In honor of The Old New Thing I call these “Vogon vulnerabilities”: I have a marvelous exploit in mind that pwns anyone I have root access to
Why would anyone ever exclude true mitm?

Various domain registrars have been compromised over and over again (often by children!), resulting in companies like Tesla and Cloudflare getting owned.

The reality is that any vaguely competent attacker can compromise a court clerk and just compel e.g. the .com registry to hand over whatever domain they want.

Although I suppose the aforementioned problem has significant implications beyond dns…

>Why would anyone ever exclude true mitm?

Same reason security programs exclude social engineering, even though that's a pretty common way for companies to get pwned.

Excluding SE is to make sure people do not spam customer support and launch annoying phishing campaigns. None of that is applicable for local software running on your own computer.
No, excluding SE is to make sure the bounty program is incentivizing things that inform the product security team. Social engineering is a corpsec function; they're not even the same teams.
Out of scope does not necessarily mean out of impact. It is merely a question of how far a company wants to be responsible for the environment their software is run in. Most of the time that answer is "not much."
Out of scope in this case means "we don't wanna pay you"
Apparently it also means "We don't want to pay our engineers to fix this".
But I use a Wi-Fi password, so my phone says it's secure!
AMD's inability to make good software has been a recurring problem for decades. Many years ago I had some success with their optimising compiler, but everything else I've touched was bad. A real pity.
Yes, their software is terrible across CPUs and GPUs, and continues to be. So many trivial bugs just never fixed.

It has literally cost them a Trillion dollars in market cap - Nvidia's CUDA is a big reason they're so much bigger than AMD.

And that’s saying something, because the CUDA stack is a PITA.
AMD somehow got success, but their company culture and pay is shit. They expect PHD level experience but expect pay like peanuts....
> expect PHD level experience but expect pay like peanuts

Thought this was par for the course in closer-to-hardware engineering.

Never understood why the objectively way harder jobs pay so much worse as an industry.

Their pay is shit. I interviewed with them 3 years ago and they offered me peanuts I rejected their offer.
The "signature verification" in the fix being CRC32 is pretty hilariously clueless.
It's technically possible (though I don't know if they actually do this) that they're not referring to a signature check in the download part, but are verifying the code signing signature of the executable downloaded. You'd only notice the CRC if you were looking at the downloaded content, but if the updater refuses to launch an executable that isn't signed by AMD's cert then they would be fine.

Given the way AMD has been treating this issue, I'm assuming they're just incompetent, though.

The article has a screenshot of the decompiled code showing that they're just running the downloaded executable immediately, without any additional checks on the content.
A manager somewhere made the embarrassingly wrong decision to not fix this, and they’re too egotistical to correct their mistake.

That’s my take.

Especially because if they had read about or studied this problem they would find tons of prior art where CRC32 was considered not secure for solving the problem. CRC32 solves a different problem -- how do you verify that the data that was received is identical to the data that was sent. It makes no guarantees about who is sending the data, which is the real problem signatures solve.
More specifically, it solves the problem of verifying that the data received was not accidentally corrupted somehow. Unlike cryptographic hashes, CRC32 does not do much to defend against deliberate, malicious modification. It's too easy to craft some different data that matches a given CRC32 value.
Computing a CRC is equivalent to attacking it. The checksum is the value that produces a certain fixed constant when appended to the data. This is why you'll often see checksums as the last field in a message. It allows for hardware to verify the entire message by checking if the CRC of the bytes equals that fixed constant without having to parse it.
They should have done base64 encryption before the crc32. noobs
Well that explains why I get that random console pop up from time to time, thanks for the insight into what's going on!
AMD didn't deny it was a vulnerability; they denied it was in the scope of the bounty program.

Remember that at giant tech companies, the incentive is to pay out bounties --- there are people on the vendor's team whose performance is measured in part by how much the program pays out.

What hair is this splitting? The issue was that AMD allowed a known and serious security vulnerability to exist within their customers’ systems, for months, and acted with a lack of candor while doing so.
It's not hair-splitting; it's central to the idea of a bug bounty. Too many people have weird ideas about what bug bounties are for.
Okay, fair. I was thinking mostly about the high-impact issue of preserving the security vulnerability and how an essential vendor was not being candid, but you are also right to note how AMD was avoiding its responsibilities to the individual researcher himself.
I mean I think you think you're doing bank-shot snark here, but what you're really revealing is that your premises hinge on AMD trying to get out of paying a bounty simply to avoid paying it. Since we know up front that's not one of AMD's incentives, what does that do to your argument? It can't help.
Yeah, like the weird idea that those programs are intended to in some way reduce the number of exploitable bugs actually out there.
That's in fact often not their core purpose!
What is it?
... which is why the rest of us should give them, and those who operate them, zero respect.

Nobody but AMD gives a fuck about AMD's internal policies or motivations.

They wanted to keep it quiet. As if they did not mind if it was exploited by those with access to international network links.
The discussion the video references [1]

[1] - https://news.ycombinator.com/item?id=46906947

The original post [1] now includes an update:

  UPDATE! Within a day of this blowing up on Hacker News, AMD reached back 
  out to me and said they would be looking into the matter after all.
[1] https://mrbruh.com/amd2/
> In my frustration, I decided to punish this software

Love this. I am frustrated by idiot software features everywhere, but am not triggered yet to punish them. AI automation is coming close however.

I got so mad at plex/jellyfin's crap I vibe coded an entire entertainment system out of spite.

Works great!

> 124 days to get AMD to add an s to a couple of HTTP URLs!

I disagree that they should only add HTTPS and call it done. They should also add some kind of signing check before running the payload.

If anything I'd say HTTPS is optional if they do that part.

They added CRC32 lol
AMD's utter incompetence when it comes to the software side of things is truly, truly baffling to me. It's not like you need a mountain of developers, a team or two on the right project would do wonders for their market share.

For example: Implement the CUDA. CUDA's won, hands down, that toothpaste is solidly outside the tube. Luckily, to the outside observer CUDA is just an API, and API's aren't copyrightable. Literally nothing is stopping AMD from hiring a relatively small team of developers to make AMD GPUs CUDA-compatible.

AMD's entire software development strategy is insane. OpenCL was doing reasonably well, and then AMD have just fully dropped support for some reason. For the - albeit not huge - but actual cross platform API that people were using to develop for their GPUs. For a while, a few cross platform tools had OpenCL backends, but nobody has been able to get AMD to fix any of the damn bugs. In my testing, bug latency for even the most trivial but important bugfixes is often 4+ years, which is utterly mad. Some parts of their compute stack is so broken its clear that nobody has ever used it. There are exploitable privilege escalation vulnerabilities caused by threaded race conditions that are wontfix

They could support OpenCL 3.0. Nvidia do. AMD just chooses not to, even though they're the ones that desperately needs to support it most

Instead, we got ROCm which has been a disaster from start to end. It barely supports windows or consumer GPUs, for some reason. Its a buggy mess, for some reason. HIP/ROCm has worse performance than OpenCL, because they downgraded their compiler and stopped extracting read/write information on variables leading to a massive loss of parallelism and utilisation on their GPUs.. for some reason. Why? What are they doing? How is this so rubbish?

Literally ALL of this is WONTFIX, and I don't have a clue why. I've filed bugs, was part of their vanguard supporter program, have tried to reach out to AMD people to (gently) explain why good support is important. Or even just figure out what technology they're even intending to support for GPU development. Is ROCm deprecated? What should we be using on windows for GPU compute on consumer hardware AMD? For the love of god amd I want to make you money

As of 2026, the best cross platform cross vendor API for doing GPU compute is.. drumroll.. OpenCL 1.2. Vulkan is getting there, but its still missing a bunch of stuff. And this is literally AMDs direct fault at this point

likewise. i'm bewildered throughout the years.

my suspicion is that it is the company culture: the hardware engineers are the real engineers. software is a triviality left for the lesser minds. the consequence is they mess up every product... everything they do needs software.

The argument I have read here on HN, is that CUDA is made for NVidia hardware, and the AMD hardware is not the best fit.

Essentially it forces AMD to play by NVidias rules, exactly like how they were forced to follow Intel rules. (Ignore for a second that the API / ISA boundary is different.)

But despite that, I also believe AMD would be better off just implementing CUDA.

They did, apparently, at one point pay someone to build that glue, and then threw it out and wouldn't let the author release it so he's been reimplementing it out of...spite? Burning desire? Unclear. [1]

I can't imagine the logic involved in "this is implemented, let's toss it in the dumpster" for that.

[1] - https://vosen.github.io/ZLUDA/blog/zludas-third-life/

HIP tries to be like this, almost API compatible with CUDA such that you just need to do find and replace. I think they even had a script to do this for you.

But the issue remains that the actual support and debugging tools remain so atrocious that it doesn't help to combat the CUDA monopoly. They've further burned a lot of trust by never really delivering on their promises to do better unless you're a customer large enough to get personalized attention from their engineers.

This ends up being a double whammy because not only are you pushing away smaller businesses, you're also pushing away single developers that go on to influence purchasing/development decisions.

HIP was such a self-own and clear demonstration of AMD software capabilities... well, the lack of software capabilities. HIP was hard-coded for one GPU architecture. CUDA did it right, it has a intermediate virtual assembly PTX and driver compiles it to whatever actual instruction set card actually uses.

Imagine a meeting where they signed off on that. So each developer will have to provide a different binary for each of our architectures? Yep. And once we release the new architecture, developer will have to recompile his program for the new architecture? Yep. Sounds good to me.

Yes, that was in part why they've had such a terrible history with GPU support.

They lost me as a customer when they rushed dropping support for the Radeon VII because of the need to ship binaries for every ISA, and didn't deliver proper 5700XT support until it was outdated.

AMD, just pay the man.

You want this stuff disclosed to you.

Actual write-up rather than overwrought YouTube drama: https://mrbruh.com/amd2/

A non-default-installation set of AMD tools (Ryzen Master and probably others) had an auto-updater which used HTTP instead of HTTPS. It's clear this is a feature they'd basically forgotten about; it even pointed to an ATI domain. A third-party bug bounty company rejected it because MITM was out of scope. AMD are incompetent at making software (news at 11), kept asking for extensions, and took an incredible amount of time to deal with it. Eventually they removed this updater entirely and replaced it with one in the app (rather than the installer) that uses HTTPS + a CRC32 (for some reason). The initial vuln was very stupid and should have been fixed faster. As for the current system, if you're mad about HTTPS-protected auto-updaters (which is valid), you've probably got a lot of them to go to war against.

> If you are an AMD user...

Don't bother to use Windows?

There's two requests involved for the auto updater, one to grab the XML file, and one to grab the driver file over plain http.

If the autoupdater can't handle the redirection when grabbing the XML file, then it's a case of accidental safety by mistake that would prevent grabbing the plain http file.

Thank you for looking into this, I also have the annoying pop-up and have been suspicious of it…
Congratulations, you found the government backdoor!
Glowing!
Jesus after reading this I'm wondering if I want to switch my AMD Framework 13 Pro order to Intel, but the IME runs on Java if memory serves and it's not like they'll be any more secure.

I just hope we can get Libreboot working with Framework sometime.

Multi billion dollar company, by the way.
I say the same when I dealt with Amazon's website, and to a smaller extent eBay.com. Don't forget Facebook Marketplace.
Let this lesson be learned: "In my frustration, I decided to punish this software by decompiling it to figure out how it worked,"

Note to self: Never piss off a programmer, while he is gaming. To quote: "You're gonna die for that." -Duke Nukem.

I think we can all agree that MiTM is a valid attack vector and this should have paid out the bounty. AMD won't do it, but perhaps we can crowdsource it - the dude deserves it. Join me in doing this: https://ko-fi.com/mrbruhh (identical link to the one in the write up, feel free to verify).

I started it with $100 - https://ko-fi.com/transactions/03df753c-09b0-4972-8e53-adf06...

Seems like white hat work is pretty fruitless nowadays. Disappointing.
They keep choosing to work whitehat instead of blackhat, which is all AMD ever wanted.
AMD software is often utter trash.

I am a diehard fanboy of their GPUs, and have been since they were still ATI but I had to finally purchase an nvidia GPU because of how bad AMDs software quality is.

My powerful 5700XT spent two years basically broken, because the default, driver provided fan curve locked the fan at 27%. For two years, I couldn't figure out why my GPU constantly crashed, because it was overheating, because the default fan curve prevented the GPU from keeping itself cool and it would eventually just give up.

That diagnoses was complicated by the fact that AMD GPUs just resetting is very common. There's a watchdog timer in Windows that resets parts of the GPU stack because Microsoft is traumatized by 60% of Windows Vista BSODs being caused by bad nvidia drivers. Apparently sometimes if you increase this watchdog timer, the GPU eventually finishes whatever was giving it trouble.

But I still love AMD, and the ryzen line is a great value in the mid range. So I bought another AMD CPU and am very happy with it. But it somehow included software and this specific auto updater utility. Which I don't need, since I don't want to update the drivers for a GPU that I shouldn't be using (maybe except some video encoding lift, but my GPU can do that too). But I could not figure out a way to kill or prevent this stupid little autoupdater utility which always steals focus, for no reason at all. It shouldn't even be popping up a CLI! Windows task scheduling is incredible and would do this without a problem, and give you all the infrastructure to notice this was happening!

Drivers got better after ATI merged/got bought by AMD, but ATI has a loooooong legacy of terrible drivers in Windows.

The funny thing is, in Linux, the drivers are pretty great as far as I can tell. It's not like there aren't bugs, probably, but mostly everything "just works". You can't depend on FSR in Linux, for example - Doom Eternal just goes blank if you turn it on. I can live without it, though, and everything else seems fine, including performance.

Nvidia linux drivers make me quite upset - they're fine once you finally get them working, but you approach Nvidia driver updates with extreme caution in Linux

Gaslighting does not mean lying.
Yeah, it's annoying. But it's been captured by popular culture as meaning a blatant lie - one where the liar knows the truth is or was available/obvious. A "don't piss on my leg and tell me it's raining" lie.

Or, alternatively, and especially in gender relations, any lie intended to manipulate or demean another person. As opposed to lying to protect yourself, to swindle somebody, or some other reason. This is closer to the original idea, but still not there.

Such a bug could have been exploited by certain big state actors.

Those that have access to international network links.

Those that have the ability to generate new firmware that simply passes the CRC32 checksum.

A bug in a nonfunctional autoupdater. Big state actors. Got it.