AMD told MrBruh that all update communications now use HTTPS and that updates undergo signature verification. The researcher says he verified the HTTPS claim, but found only a CRC32 check on the downloaded executable, which is not considered a cryptographic signature.
This is the most shocking part. You’d think that AMD as a high-tech company has some smart people working for them. These are very basic things that any half decent programmers should get right. If at no part of the process of implementing this anyone brought up that this is not secure, that is extremely worrying and indicative of a very broken development process. It’s not like a proper cryptographic signature costs extra. This is just pure incompetence.
Not surprising at all. I work in IT and security is by and large reactionary and based on scans that are often rudimentary. As far as training devs on good security practices there’s next to nothing. You learn from getting your hand slapped or you don’t learn at all.
Well the next time someone finds a bug in their software they will have to find other ways to monetize it.
AMD has always sucked at making software. The reason why NVidia gained the AI market is because NVidia worked to write and support all the CUDA libraries. AMD devs are so bad they even struggle to just replicate the APIs NVidia already designed year earlier (ROCm/HIP projects). Even Intel who arrived much later almost managed to catch up with their own HW/SW stack (I think they gave up afterward).
The problem with using CRC32 is it reversible and has high collusion rate. An attacker can easily make a file the generates the same hash. This tool a few minutes of searching online. It appears that people who work at AMD don’t even know how to do proper research. All they have to do is look up how to make a secure updating process.
The problem is that a CRC32 checksum is not a signature. Doesn’t matter if they use the most complex checksum in the world or not, what they need here is a signature
What does it matter if it’s CRC or sha512 if they are using an unsecured connection to transmit them? A stranger who has already acquired capability to modify the payload in transit can also modify the checksum. A better hash will not solve this problem.
They use https now, but use CRC for signature verification:
AMD told MrBruh that all update communications now use HTTPS and that updates undergo signature verification. The researcher says he verified the HTTPS claim, but found only a CRC32 check on the downloaded executable, which is not considered a cryptographic signature.
I could be wrong here, but I believe they should use a combination of SHA256 and PGP for signature verification.
Oh, okay, so maybe I misread the sentence. I thought the implication was they used crc32 as opposed to HTTPS. Not sure why you need an additional layer in addition to https- as long as the certificate chain is setup properly. And again, you’re not gaining additional security if you submit the hash (or a gpg key) through the same channel. So if they already use https and just want to check for broken downloads, crc32 is perfectly fine. It’s just security theater at that point.
An attacker can still send a compromised payload if there’s no signature verification of the update. It takes a more sophisticated attack (e.g. supply chain attack, hijacking AMD’s update website, etc.) but it has happened before to other companies. If the payload is signed and verified, an attacker would also need to gain access to AMD’s private key to successfully send out a bad update. Assuming reasonable security, getting that private key would be a lot harder to get on top of somehow compromising AMD’s update web service.
Also CRC checks over the internet are sort of silly and redundant since every packet sent would already be subject to a similar CRC check and bad packets would be ignored (dropped and re-requested). It would only prevent corruption on disk or in memory which are a lot less likely than transmission corruption.
This is the most shocking part. You’d think that AMD as a high-tech company has some smart people working for them. These are very basic things that any half decent programmers should get right. If at no part of the process of implementing this anyone brought up that this is not secure, that is extremely worrying and indicative of a very broken development process. It’s not like a proper cryptographic signature costs extra. This is just pure incompetence.
The very smart people working on their architecture and chip design are very much not the same people who are working on their desktop software.
Not surprising at all. I work in IT and security is by and large reactionary and based on scans that are often rudimentary. As far as training devs on good security practices there’s next to nothing. You learn from getting your hand slapped or you don’t learn at all.
As someone who is frequently the one slapping hands (and backs of heads), I can confirm this.
And still they don’t learn.
AMD has always sucked at making software. The reason why NVidia gained the AI market is because NVidia worked to write and support all the CUDA libraries. AMD devs are so bad they even struggle to just replicate the APIs NVidia already designed year earlier (ROCm/HIP projects). Even Intel who arrived much later almost managed to catch up with their own HW/SW stack (I think they gave up afterward).
The problem with using CRC32 is it reversible and has high collusion rate. An attacker can easily make a file the generates the same hash. This tool a few minutes of searching online. It appears that people who work at AMD don’t even know how to do proper research. All they have to do is look up how to make a secure updating process.
The problem is that a CRC32 checksum is not a signature. Doesn’t matter if they use the most complex checksum in the world or not, what they need here is a signature
Day by day, vibe-code all the way…
What does it matter if it’s CRC or sha512 if they are using an unsecured connection to transmit them? A stranger who has already acquired capability to modify the payload in transit can also modify the checksum. A better hash will not solve this problem.
They use https now, but use CRC for signature verification:
AMD told MrBruh that all update communications now use HTTPS and that updates undergo signature verification. The researcher says he verified the HTTPS claim, but found only a CRC32 check on the downloaded executable, which is not considered a cryptographic signature.
I could be wrong here, but I believe they should use a combination of SHA256 and PGP for signature verification.
Oh, okay, so maybe I misread the sentence. I thought the implication was they used crc32 as opposed to HTTPS. Not sure why you need an additional layer in addition to https- as long as the certificate chain is setup properly. And again, you’re not gaining additional security if you submit the hash (or a gpg key) through the same channel. So if they already use https and just want to check for broken downloads, crc32 is perfectly fine. It’s just security theater at that point.
An attacker can still send a compromised payload if there’s no signature verification of the update. It takes a more sophisticated attack (e.g. supply chain attack, hijacking AMD’s update website, etc.) but it has happened before to other companies. If the payload is signed and verified, an attacker would also need to gain access to AMD’s private key to successfully send out a bad update. Assuming reasonable security, getting that private key would be a lot harder to get on top of somehow compromising AMD’s update web service.
Also CRC checks over the internet are sort of silly and redundant since every packet sent would already be subject to a similar CRC check and bad packets would be ignored (dropped and re-requested). It would only prevent corruption on disk or in memory which are a lot less likely than transmission corruption.