Improving the world’s security at Google. Opinions are mine.
Zurich, Switzerland
Joined September 2014
- Tweets1.2K
- Following284
- Followers11.3K
- Likes924
Introducing PageBreak, an agentic vulnerability discovery tool using deterministic validation to achieve a near-zero false positive rate.
How it works: blog.google/security/agentic…
For real-world vulnerabilities PageBreak has discovered 👇
bughunters.google.com/blog/p…
We just published two blog posts about PageBreak, Google's AI web security scanner (that I've helped develop for the better part of the year). We share our approach and a couple of findings, enjoy!
- blog.google/security/agentic…
- bughunters.google.com/blog/p…
@[email protected] retweeted
“And no, AI isn't perfect. But Christ, anybody who points to the problems at AI had better be looking in the mirror and pointing at themselves at the same time.
Because it's not like natural intelligence is always all that great either.”
lore.kernel.org/linux-media/…
Thanks for the appreciation!
In theory this could be findable by just reading the HTML spec but I assume that would be difficult for agents. But if they can fiddle with the HTML and fuzz it, I’d expect them to be able to discover it.
I remember stumbling upon that when doing my copy&paste research a few years ago. AFAIR there was something similar in one of macOS built in apps but I don’t remember which one.
This is a really awesome summary of how to approach bypassing HTML sanitizers, I highly recommend to check this out!
DOMPurifyバイパスの歴史をまとめたページ。HTMLサニタイザーバイパスのノウハウがぎっしり詰まってる。
特に今年に入ってから発見された<selectedcontent>のバイパスは驚いた
github.com/cure53/DOMPurify/…
@[email protected] retweeted
Have been working on agentic vulnerability discovery for a while now, happy to share about CodeMender! 🚀🚀
While AI is helping us tackle big challenges, it also poses new potential security risks. That’s why we’re focused on building tools that keep the broader ecosystem safe — like CodeMender, a code security agent that automatically finds and fixes critical software vulnerabilities.
Today, we’re inviting a select group of experts to test a new CodeMender API, and we’ll be launching it more broadly soon.
#GoogleIO
ALT Text reads "CodeMender" over an abstract pink and blue grid illustration.
Anyone I know interested in joining the Google Security Team in Zurich? Let me know, I can give a referral :D
Here's the job posting: google.com/about/careers/app…
Cool research and nice write up! Also thanks for the shout outs!
Side note: I’m actually planning to propose yet another spec change this year to try to kill even more mXSS-es so enjoy them while you can 😅
That was a nice bug, thanks for the shoutout!
2️⃣ XSS in GMail's AMP4Email via DOM Clobbering
Michał Bentkowski (@SecurityMB) exploited DOM clobbering to achieve XSS in Gmail's AMP4Email feature. Found that AMP4Email allowed id attributes, which could be leveraged to overwrite JavaScript variables and bypass Google's strict security filters.
research.securitum.com/xss-i…
@[email protected] retweeted
Why the "Agents Rule of Two" is flawed
shhnjk.substack.com/p/why-ag…
@gridogram I forgot to play gridogram yesterday. Is there an option to play yesterday's puzzle?
@[email protected] retweeted
Cross-Site ETag Length Leak
blog.arkark.dev/2025/12/26/e…
I just posted the author writeup for impossible-leak in SECCON CTF 14 Quals. As far as I know, this is a new XS-Leak technique! The ETag header can become a side channel :)
Ah, it's probably from the old leak: haveibeenpwned.com/Breach/Ca…
Has there been some email leak from @Canva? Just got a spam message for an email used only for Canva.
@[email protected] retweeted
We launched a redesigned Project Zero website today at projectzero.google !
To mark the occasion, we released some older posts that never quite made it out of drafts.
Enjoy!
Interested in the security of AI Agents 💁🛡️?
Then you've likely heard of "prompt injection", but do you know what "task injection" is? If you're curious, check out our latest post for a description and some real-world examples we discovered.
bughunters.google.com/blog/4… bughunters.google.com/blog/4…
@[email protected] retweeted
my new blogpost is out!!
this one talks about a new web vulnerability class i discovered that allows for complex interactive cross-origin attacks and data exfiltration
and i've already used it to get a google docs bounty ^^
have fun <3
lyra.horse/blog/2025/12/svg-…
Exploiting bentkowski.info/2022/11/goog…
