Leandro Barragan

I agree with folks this who say that this year will be an absolute deluge of CVEs found with AI. But I also worry that it will reveal the limits of the "we'll just fuzz out all the bugs" mindset

This week, Disclosed. #BugBounty H1-65 Singapore & H1-468 Stockholm winners, new H1-Elites, Google’s AI VRP, YesWeHack wins EU tender, new programs, tools, write-ups & videos — and more. Full issue → getDisclosed.com Highlights below 👇 @tiktok_us & @okx H1-65 (Singapore) winners: MVH — @corraldev; Community Choice — @Agornello; Best Collab — @kevin_mizu, @infosec_au, @hash_kitten & @HackerOn2Wheels, @ledz1996. @Hacker0x01 H1-468 (Stockholm) winners: @Blaklis_, @snorlhax, @DoomerOutrun (MVH & Best Collab); @holyfield (Eliminator); @Rhynorater (Eradicator/Exterminator); @joaxcar (Community Choice); @alicanact60 (Epic Unreal Hacker). New @Hacker0x01 H1-Elites for 2025: @niemand_sec, @ArchAngelDDay, @mallocsys, @alicanact60, @_godiego_ — congrats! @busf4ctor & @monkehack take AI Bug Research honors at Google VRP Mexico. @yeswehack wins the European Commission’s 4-year bug bounty tender to secure open-source assets. @Hacker0x01 paid $81M+ in bounties last year — AI vulns spiking. @immunefi rolls out new anti-spam rules (Oct 1) @Bugcrowd opens @SimpliSafe program (up to $6K) @TomKuCoin launches KuCoin program (up to $15K). Google launches a dedicated AI Vulnerability Reward Program (up to $30K) to clarify the scope of AI security findings. Cloud Software Group / @NetScaler goes public with a bug bounty on @Hacker0x01 . CTFs & events: @hackthebox_eu x @Hacker0x01 AI Red Teaming CTF (500+ participants) @bugcrowd Mind Cathedral (50+ teams, 300+ submissions) Videos and write-ups from @NahamSec, @amrelsagaei, @ctbbpodcast , and more. New tools: graphql-cop, HTML-Search-Engine Chrome extension, file_upload_payloads repo, Gemini-API-Key-Exposure-Scanner — handy for recon & CI/CD testing. Notable writeups & research: RCE guides, Next.js testing, supply-chain attack techniques by @0xLupin, SSRF/XSS escalation threads, and a leak exposing personal info of Oscar nominees by @galnagli. Full links, winners, writeups & tools → getDisclosed.com The bug bounty world, curated.

It's out!! You can now watch @djurado's and @niemand_sec talk: "Prompt. Scan. Exploit - Ai's Journey Through Zero-Days, and a Thousand Bugs". Learn more about @Xbow and autonomous hacking. You can watch it in our Youtube channel exclusively: youtu.be/y_aQQmDMaY4. Enjoy!

Enterprise security products don’t need to be secure (or even good at all) to be sold like hotcakes. 61B market cap and a myriad of vulns. No one cares about that other than people like us, this is as old as time :(

I discovered how to use CSS to steal attribute data without selectors and stylesheet imports! This means you can now exploit CSS injection via style attributes! Learn how below: portswigger.net/research/inlin…

@moyix @ESETresearch It also uses quite a lot of ram, I’m curious why the dev chose this model

@ESETresearch I guess if it’s just getting ollama to do the model download?

#ESETResearch has discovered the first known AI-powered ransomware, which we named #PromptLock. The PromptLock malware uses the gpt-oss:20b model from OpenAI locally via the Ollama API to generate malicious Lua scripts on the fly, which it then executes 1/6

Legba v1.1.0 is out! 🥳This is a major release that required a significant amount of (human) effort, bringing several key improvements that deserve individual attention. 🧵👇

@pxmme1337 @rez0__ For example, here we explain why we don't have to deal with false positives (it would have been horrible to validate so many runs manually): cdn.prod.website-files.com/686c11d5bee015…

@pxmme1337 @rez0__ Xbow was never a product designed to do bounties and we were clear about that. It’s an autonomous pentester, h1 was basically a playground to test our capabilities. There is no need for speculations when we were quite transparent about most of the topics you mention.

In the post: > XBOW isn’t here to replace pentesters or researchers Yes it is. And that's okay. Cars replaced horses. > it augments teams. By removing routine burdens from penetration testers, it frees them to explore frontier vulnerability classes and the application-specific bugs that matter most. I DO think this is also true. Hackbots will find X% of bugs (and that number will continue to go up) but hackbot+smart humans will find X% + Y% of bugs where Y represents the delta that a smart operater adds. The smart operator intelligently drives the hackbot which means less token cost and smarter decisions. If XBOW actually intends to augment teams, the best way would be to offer a Caido plugin with something like a pay-as-you-go pricing tier to give the power of their system to smart operators.

I had the pleasure of working at the company this genius founded in 1996 (!). He and a handful of others shaped the spirit of the Argentinian hacking scene, sharing their knowledge and infecting us with curiosity.

@galnagli You can even replace me with an MLP neural network from the 80s 😜 so no need for an LLM. You can even argue I may have been _hindering_ xbow development all this time!

@lean0x2f Awesome! I was worried about your job for a second 😅 when it replaces you I'll start panicking 😉

"XBOW isn’t here to replace pentesters or researchers; it augments teams. By removing routine burdens from penetration testers, it frees them to explore frontier vulnerability classes and the application-specific bugs that matter most." xbow.com/blog/xbow-on-h…

@rez0__ btw you wouldn't believe the number of issues we got in PPR/New from weeks /months ago🙃 so we'll be continue receiving points for a long time.

But it also says this: > While we are no longer focused on climbing the leaderboard, this doesn't mean we're leaving HackerOne. We will continue to use it as the place where we work with the community to gather feedback and to battle-test our most experimental capabilities before they touch a customer environment.

Xbow concludes its HackerOne & Bug Bounty efforts. It was a nice playground to hack live, real-world targets. Our pentest customers are already benefitting from all the experience we harvested :)

@rez0__ Heck, they can even fold proteins. My point is that at least I don't have to read those horrible plaintext RFCs 🤣 The definition of frontier implies that it's constantly moving. If AI shifts it, great, I'll focus my real neurons somewhere else.

@lean0x2f mostly agree but frontier research is often: - testing a bunch of stuff where assumptions are made by other testers and existing tech - comparing implementation of RFCs against actual RFCs. and both of those are things that current models with agentic wrappings can do well.

@rez0__ Yeah of course, we'll continue to use the platform to test experimental features (but only for those). We have our benchmarks, but HackerOne is the most realistic playground for Xbow.

@galnagli @kuzushi @Xbow That is, the detection part is not limited in any way, other to what the models know and what we teached it to do. It’s non deterministic, that’s the idea. The validation part though, to filter out false positives, is what you are seeing there.

@galnagli @kuzushi @Xbow Ugh I hate this UI, we should change it. During black hat I had to pause and explain this part all the time. Those are not bug classes but instead the validators we use. So, it will find for example an XXE or a deserialization bug, and it will validate it using LFI for example

No, it wouldn't. If you could have, you would have. Entire companies were built around that before ai was introduced, and it didn't scale proportionally.

Lot of people asked me about the models XBOW is using. This and Albert's blogpost about alloys may answer some of your questions (alloys here: xbow.com/blog/alloy-age…)