Ian Klatzco

hacker 🌠 @sigpwny prev: @leviathansec and @SPR_Illinois 🦋

[Ian Klatzco @ian5v]

@richinseattle I'm particularly interested in the new-ish secure enclave feature, "NVIDIA Confidential Compute".

[Ian Klatzco @ian5v]

Hey @richinseattle! Thanks again for your Toorcamp talk last year, I really enjoyed getting curious about e.g. why .safetensors is "safer" than .pickle.

Do you know anyone who is really into GPU security? I've been curious about their fuzzability for a while.

[Ian Klatzco @ian5v]

update: they enable you to know that the request you're seeing Now was signed by the same secure element you saw Earlier.

(now, there is also syncing between devices in an eg apple/google ecosystem)

[Ian Klatzco @ian5v]

why on earth are passkeys better?

they seem harder to understand.

the only benefit seems to be that the sign-in flow is slightly smoother in some cases.

[Tim Urban @waitbutwhy]

Good day to remember just how big that green tree is

[Ian Klatzco @ian5v]

@TaroNuke fraud is expensivvve

[Ian Klatzco @ian5v]

sdv3 might be a transformer (cf sdv2 which is a vae+unet)

[Ian Klatzco @ian5v]

USENIX Security ML / AI Track

[Ian Klatzco @ian5v]

usenix dot org really needs to put the papers in a cdn....

[Ian Klatzco @ian5v]

input: manually labeled dataset of unsafe concepts (gun violence, banned flags, sexual harassment)...

result/contribution: UnsafeConcepts dataset of 75 concepts / 1500 images in 9 categories

[Ian Klatzco @ian5v]

identifying unsafe content using vision language models