Making QUIC Quicker with NIC Offload

What do you mean by privacy issues with QUIC? As I understand it, it’s just another application layer protocol.

Iirc, Brave browser began disabling QUIC because the connected server was able to fingerprint the connecting party, and they deemed it an intentional part of the protocol design. Essentially connecting anonymously to a server is not possible. I'll try and dig up the blog post.

Yes, it's an attractive place to add surveillance code for all NICs. But not all NICs take over encryption/decryption. And if other components do the encryption/decryption, NICs don't see plaintext - just (lots of) metadata.

If/when they take over encryption/decryption functions, that would make them an even better place for hard-to-detect, user-disloyal surveillance code.

> If that software then exfiltrated valuable info – session keys, or a digest of hotwords seen, or every string typed into targeted HTML form fields – how would you know?

How does this magical exploit work, exactly? How would it extract keywords from forms, when forms are ridiculously dynamic per-website, per user, and per form? (Many forms aren't even actually that, but text boxes that are fed directly by JavaScript event handlers into a Websocket, these days.) How do they target these groups using and installing these the NICs, and ensure they're all put in the right place? Do they literally backdoor every device that exists? No meaningful backdoors were ever found in Supermicro devices, for instance, despite immense pressure to find them. The same with session keys, what "session keys" are being compromised? Session keys exist at the TLS level, the application level, and conceptually many others. TLS keys already aren't useful for anything beyond their used connections for instance, and flow identification can vary in many ways, so how would they correlate user-to-flow in order to target attacks? If they're not doing targeted attacks, then what are they doing?

How is it going to exfiltrate that data if the NIC is on an isolated network (e.g. in some mesh topology, because it's only used intra-rack)? How do they plan for that, by exploiting the software stack? If they can exploit the software stack already, why not just compromise the host? You're telling me NSA spooks can in advance make sure your datacenter rack is organized so that RF signals can be used to wirelessly transport data on the spectrum and nobody has ever had proof of this, but they can't just, you know, write a software exploit? Issue a gag order? And that they use this capability with some regularity? If it wasn't regular, why do it in the first place when the risk of discovery is higher? How do they predict these ridiculously variable conditions in advance, and how are they not discovered? And above all, why would all this be something you immediately jump to, instead of just, say, normal infosec style attacks? You can literally buy 0days for COTS software, on the market. Your original post was motivated by fear of some Chinese person working for the Chinese government. Why? Because China is a totalitarian government but also they somehow need university students to make backdoor'd NICs, in conjunction with researchers from Europe who wouldn't, like, detect them? The ruling party controls the entire military and governs the internet and can already decrypt all connections, but they need this too?

This kind of stuff is just sci-fi/mil-tech fantasy scenarios. We can all hash out incredibly-specific ridiculous infosec attacks all day that we could pull off in a controlled environment, but that doesn't mean any of them are actually more useful or viable than others. People on this site desperately want to believe the world is filled with action movie sci-fi hacker scenarios that can just be defeated in a technical one-up kind of way, because it means the solution is purely a technical problem, rather than the real alternative: you will get thrown in a dark jail cell for 500 years without appeal if you don't comply with this order to install a backdoor directly in your datacenter.

This one?

https://content.sciendo.com/configurable/contentpage/journal...