You don’t need to (and should not) forward your SSH agent just to set up a tunnel, especially with the default OpenSSH ssh-agent. It allows the intermediate host to decide what operations to do with your key(s). `ProxyJump` is a way safer default.
Password gets sent to the server. Private key does not get sent to the server, and authentication protocol can be forwarded safely, enabling things like ssh-agent, FIDO SSH keys. This is solvable with passwords (see: augmented PAKE), but `PasswordAuthentication` isn’t that. But there are also settings where none of this really matters, yes, to be fair if that’s what you’re referring to.
> password doesn’t get sent to the server, hash does.
Kind of irrelevant what gets sent to the server and what happens on the server.
Password is already easily defeatable long before it even leaves your computer, thanks to keyloggers and/or password manager breaches and/or old-school shoulder surfing (assisted in the 2026 era by covert filming and AI).
All of which are not vulnerabilities with hardware-backed keys which you don't even need a Yubikey for these days since most modern computers and smartphones have secure enclaves and/or TPMs. (Yubikeys are still great though).
I’m curious what both of these comments are referring to, since it’s not supposed to take any privileges to answer a question on Stack Overflow (you can even answer without making an account). Maybe some anti-spam block?
This is why you give the updated answer as a comment under the top answer, with a casual "shouldnt we do xyz in [anno domini NOW] instead". Then some established answerer takes what you just said, edits it into the top answer, and steals all credit. But hey, at least you got the answer to be updated, not in your words, and no trace of your input into it. But still updated, right?
I'm not pouring one out for SO's death. Because I already did years ago.
- but anyone who deploys thousands of lines of vibe-coded C that implements a transforming HTTP proxy from scratch instead of a patch (hey, it “might even be safer than the patch”!) deserves what they get
Both of those are coming from telemetry, and mainly show attempts. Those telemetry are generally heuristic based so they overcount and are most often unsuccessful - it took much layer for actual honeypots to be hit. See my timeline in my other comment if you’re interested in details and extract dates
Linking a reasonable way to create a necessary RAM disk rootlessly with namespaces to kernel exploits that involve namespaces seems like a massive reach.
Yeah, this particular example was pretty sketchy on Cohere's part. Unsharing a namespace like that is a pretty standard way to safely do things that would otherwise require root.
Yes, this exposes some root-level kernel features to ordinary users in a restricted way, and yes, those features sometimes have security bugs that aren't visible to users outside a namespace.
But nothing that model was described as doing seemed particularly sketchy? At least not from the information we were given? It was using a normal kernel operation that I used regularly, not trying find a kernel exploit?
So that part of the article left a bad taste in my mouth.
But that was the whole article. Their exfiltration of data to Beijing, wasn’t. And it was told to build something to monitor against exfiltration attempts, shouldn’t it try one? It wasn’t sensitive data, it wasn’t even data, it was a test string. Everything they described was very benign.
Should the model have asked for an endpoint to test against? Probably yes, but this is not a very large model and the largest frontier models regularly do things like that without asking me.
The reality of running recent Chinese open weight models is that they are all overtuned for successfully completing tasks. I can ask Qwen3.8 Flash Next to grep for regexes that use PHP-specific features that might be problematic to port to Rust. But instead of just grepping, Qwen3.8 FN will spend 30,000 tokens writing a Rust test harness, exhaustively check every regex, consult a bunch of docs, and write me a 2 page report. Q3.8 FN is deeply convinced that it's going to get extra credit for doing more than it is asked. (To be fair, it's a prototype release.)
But if your sandboxing isn't set up to handle some whacky weirdness, then you can't even trust Claude or ChatGPT. Which is important—the big vendors often do provide sandboxing or command filtering in their harnesses, but some popular local harnesses like Pi.dev are intended to rely on external sandboxing tools.
I noticed similar behavior too across various different models. So I started asking it to create implementation plan documentation for everything, even for a small fix. I review the plan, clarify and update it very carefully before allowing it to proceed. I haven't had this type of unnecessary "extra credit" efforts since then.
reply