!Manan retweetledi
0 chrome submissions(?), 3 firefox renderer, 1 edge, 2 safari renderer, and exchange/sharepoint each.
devcore still showing up with multiple submissions in the hard targets, including exchange, sharepoint, and edge, from known names.
so why are we not seeing a huge amount of submissions? i think, either:
1. defense got stronger with llms, and software like chrome/firefox is fixing a ton of bugs before they ever reach pwn2own or
2. hacking of complex software is still bottlenecked by a small number of top-tier researchers.
i would guess it’s the latter(?).
there is no denying the fact that, llms are probably closing some defense gap, but i think that doesn’t mean the asymmetry moved to the defense side and making defense stronger, i still think it’s the usual attacker-favored game.
and looking at pwn2own submissions, it seems pretty obvious to me that llms are still only as good as the operator using them.
there are only a few people good enough to point them at hard targets properly and use them to actually accelerate research.
cuz, if llms were actually giving everyone exploit superpowers to "anyone", you’d expect more people showing up with chrome/exchange/browser-class bugs.
instead, what we’re seeing is still mostly people with skin in the game hitting the hard targets.
zerodayinitiative.com/blog/2026/5/13…
s1r1us (mohan)@S1r1u5_
this year's pwn2own isn't just interesting because there will be lots of entries with AI+human. it is also interesting because a) anthropic burned a ton of tokens on firefox, basically running claude in a loop until it found something for a month, probably exhausting whatever claude can one shot. b) if someone submits full chain without much use of ai, it tells you one shotting plateaus and these models are bit like fuzzers than seasoned security reseachers. c) even if they used an llm to find the bug, this tells us scaffolding/harnesss design, prompting, and the operator matters a lot.
English
