Love the part where he claims that if your users are authenticated, it’s not untrusted input. I mean, surely you trust all of your users to run any code on your server, right?
Impressive and unsurprising. As soon as you start getting complex libraries with multiple dependencies it becomes nearly impossible to review everything. At one time I had an interest in contributing to some AI libraries, but they’re a mess as soon as you go looking for points of improvement.
Counterpoint: “not knowing your libraries” + “blind trust in the maintainer” will give you stuff like this: https://bitbucket.org/snakeyaml/snakeyaml/issues/561/cve-2022-1471-vulnerability-in
(the thread itself is worth a read. But also very impressive is the list of big players who fell for exactly this mentality)
Love the part where he claims that if your users are authenticated, it’s not untrusted input. I mean, surely you trust all of your users to run any code on your server, right?
Jesus that was one hell of a thread
I dont want to see the words “low quality tooling” ever again.
Impressive and unsurprising. As soon as you start getting complex libraries with multiple dependencies it becomes nearly impossible to review everything. At one time I had an interest in contributing to some AI libraries, but they’re a mess as soon as you go looking for points of improvement.