Pinned Tweet
If you're wanting to level up your code review skills, I would highly recommend @PentesterLab's on demand training, I've completed the training and the amount of information covered is insane and you learn techniques from an industry veteran @snyff.
pentesterlab.com/training/
π»π°πͺπ₯π΄π¦π€ retweeted
Not sure a PHP function behaves the same on every version?
One loop:
for v in 5.4 5.6 7.0 7.4 8.0 8.1 latest; do
docker run -v $PWD:/code php:$v php /code/test.php done
Very little work for a lot more assurance.
Had to post this one as an image to fit it all in. Worth the read.
#hackbot #BugBounty
Not hating on neo at all in fact it's very fast and such a great interface to work with but it's just these small edge cases that can make a huge difference. @pdiscoveryio @badcrack3r π«Ά
6
1,221
I Built a local bug bounty recon platform that runs in the background: Rust tools handle subdomains, DNS, probing and crawling, a local systemone model classifies assets and flags leads. Argus MCP lets a hackbot poll for leads, pull them and hunt for bugs while I test in Burp.
Very useful information π§ the quirky behaviour in programming languages can lead to interesting bypasses.
Same trap, two languages:
Python: os.path.join("/app", "/foo/bar") is "/foo/bar"
C#: Path.Combine("/app", "/foo/bar") is "/foo/bar"
In the happy path they do what you expect, so people never look further.
4
362
We built security tools to automate tasks with predefined rules, but they struggled with edge cases. Harnesses add reasoning, letting tools interpret results and adapt, with HITL keeping humans in control.
2
424
Without a good harness, you're just finding the same shit as everyone else: the new era of Nuclei.
A well-engineered harness is like giving your lab rat tren (Nuclei private templates).
But finding the right balance between manual hacking and your hackbot? activates god mode.
AI agents don't always follow the right design patterns. I built this product to help agents use established design patterns via MCP, helping keep codebases clean and maintainable. It's a tool, not a magic fix. Always review your code.
Check it out: aipatterns.dev/agents
Definitely worth a read π₯ you won't regret it and you will learn lots.
Great work Louis πͺ
This is amazing π the profundis team has completely refurbished there UI/UX and they always have clean data and some of the best features I've seen to date. It's one of the best Shodan alternatives, I would highly recommend giving them a try out.
#bugbounty #recon #profundis
We investigated a spam campaign running on throwaway domains, and found the company behind it with our new Pivot tool.
Its WHOIS was privacy-protected, but the company name was left in the organization field, on 208 domains.
blog.profundis.io/blog/from-β¦
1
9
899
Great work Harley! you're posting some great content for the community πͺkeep up the epic work.
Replying to @HacktronAI
@HacktronAI names a July intrusion chain into @OpenAI , @MetaBugBounty announces a public Muse bounty up to $300K, @efaav finds unsigned JWT auth in a Microsoft analytics API tied to 17 trillion records, and more.
This week, Disclosed. #BugBounty
Full issue β getdisclosed.com
Highlights below π
@MetaBugBounty takes its Muse AI-agent bounty public, up to $300K for qualifying vulns and high-impact prompt injection.
@HacktronAI documents a chained intrusion into @OpenAI, starting from an image parsing bug and ending in RCE plus downstream identity and SSO abuse.
@efaav walks through unsigned JWTs in an internal Microsoft analytics API, claim spoofing unlocked arbitrary queries, the write-up frames the blast radius as 17 trillion records.
@vercel consolidates private and OSS coverage into one public @Hacker0x01 program spanning Vercel products and open-source projects.
@Intigriti confirms @Adobeβs public bug bounty program migration, going live on September 1, 2026.
@AnthropicAI announces a public @Hacker0x01 bug bounty after running privately with invited researchers.
Anthropicβs September 2026 threat report summarizes observed Claude misuse, including multi-agent phishing and AI-augmented cyber workflows.
@NahamSec announces NahamCon βPrompt2Pwn,β a live hacking event format for 2026.
@forefy releases reburp, a Burp extension that exposes the Montoya API as a localhost OpenAPI REST interface for external scripts and agents.
@Pillar_sec reports session spoofing and SSRF in Grafana MCP, one issue enabled unauthenticated tool execution with server-side privileges.
@0xacb drops a GeoNetwork pre-auth RCE chain write-up, unauth formatter upload plus unsafe XSLT reached code execution.
@SLCyberSec details Goja JavaScript engine exploitation, out-of-bounds heap write to arbitrary read/write and full RCE.
@devploit ships geminiHunter, scanning web assets and Android apps for exposed Google Gemini API keys, plus validation checks.
@Antisy_Training publishes βExfil Everything,β a survey of real-world data theft chains against AI agents and their trust boundaries.
That's not all. Full links, writeups & more β getdisclosed.com
Bug bounty and offensive AI, curated.
1
2
614
π»π°πͺπ₯π΄π¦π€ retweeted
Why is AI code review so easy to adopt? What makes pentesting agents harder to deploy? Which skills should you keep developing?
Watch the recording of our webinar: "The Uneven Impact of AI on Offensive Security" with @PentesterLab's founder @snyff:
youtube.com/watch?v=IqghiUqxβ¦
Check out PentesterLabs YT channel, there is some great content for everyone. @snyff has been working very hard πͺπ₯
π―π―
One of the biggest mistakes beginners make when testing an application is not reading the error messages properly.
They send a payload, get an error and conclude that it did not work.
Then they send another payload, get another error and assume it is basically the same thing.
But an error does not mean the payload did not work. It just means that an error was triggered somewhere in the execution path, and that point may be after the interesting part already happened.
Your payload may have been parsed successfully, changed the control flow, bypassed a check, reached a different function or triggered behaviour that the previous payload did not. Something later may simply have failed.
This is why small differences matter so much. A different exception, a slightly different message, a different status code or even a different location in a stack trace can tell you that you moved forward.
Security testing is often about noticing those tiny changes and asking why they happened.
Donβt just look for success or failure.
Read the error.
5
780
Small improvements day by day, fingers crossed having an extra pair of eyes can help my team find some interesting leads to hack on π€
Been building an event-driven AI recon classifier for me and the team.
One thing I've learned: test your harness constantly and keep tweaking your skills + system prompts.
Models already know the common stuff. Build skills around the things they don't.
2
27
1,759


















