Pinned Tweet
I am very happy for the donation I had from @zseano I am very grateful because I am inspired by you, my friend, always a good man. Thank you for everything and I wish you many good things, be well my friend and thank you very much. I was very happy.
#bugbounty #twitch
👑 OFJAAAH → ofjaaah.com retweeted
Simon Koeck (@soeckly) earned $90,000 from Google after finding a DevSite vulnerability that let him run commands on Google's servers.
The attack came down to 3 security mistakes that chained together.
1. A secret exposed in a session cookie
Google's DevSite stored a pickled OAuth object inside a Flask session cookie, including its `client_secret`.
The cookie was signed, but not encrypted. That means users couldn't modify it without invalidating the signature, but they could still decode and read its contents.
So the researcher decoded the cookie and extracted the `client_secret`.
2. The same secret was used to sign cookies
Normally, servers use a private key to sign session cookies so they can detect tampering.
But DevSite used that exposed `client_secret` to derive its cookie-signing key.
With the secret in hand, the researcher could now forge his own session cookies with valid signatures that Google's servers would accept.
3. Google used Python's `pickle` to deserialize cookie data
`pickle` lets Python save objects as data and reconstruct them later. The problem is that loading malicious pickle data can execute arbitrary code.
DevSite used `pickle` to restore an OAuth flow object from the session cookie.
Since the researcher could forge valid cookies, he could replace the original pickled object with a malicious payload.
Here's how the full attack worked:
1. Decode the session cookie and extract the `client_secret`.
2. Use that secret to derive the cookie-signing key.
3. Forge a valid session cookie containing a malicious pickle payload.
4. Send the cookie to DevSite and trigger the OAuth callback.
5. DevSite validates the cookie signature, deserializes the pickle, and executes the payload.
To prove RCE, he made Google's server execute a command that delayed its response by a few seconds. The response times matched the delays.
The takeaway:
Three issues combined into one critical vulnerability:
- Secret exposure → The signing key could be derived.
- Key reuse → An attacker could forge trusted session cookies.
- Unsafe deserialization → Forged data could execute code on the server.
#BugBounty tip: Decode session cookies when testing applications. Signed doesn't mean encrypted, and sensitive information inside a cookie might be enough to compromise the mechanism protecting it.
Full write-up 👇
simonkoeck.com/writeups/goog…
👑 OFJAAAH → ofjaaah.com retweeted
Java4Shell PoC released for critical OpenJDK vulnerability that could allow remote code execution; CVE-2026-114938 PoC: github.com/ikow/java4shell-p…
Since the vulnerability has leaked: Calif discovered and reported Java4Shell (CVE-2026-114938).
We're publishing technical details by 1 PM Pacific today, after Oracle releases a patch.
Stay tuned!
👑 OFJAAAH → ofjaaah.com retweeted
Hackers não precisam de mais uma biblioteca de prompts. Precisam de skills de verdade. 💀
Conheça o NotHoly Skills.
Uma biblioteca crescente para segurança ofensiva, pentest, bug bounty, engenharia reversa, análise de malware, OSINT e muito mais.
notholy.com/pt-br/skills
👑 OFJAAAH → ofjaaah.com retweeted
Se você usa IA pra pesquisar vulnerabilidades, revisar código ou analisar binários, dá uma olhada na NotHoly. API e CLI pra trabalhar no seu fluxo, com busca web e dados de CVEs.
Promo: DeepSeek V4.1 Flash Abliterated: até 70M sendo 10M tokens/dia por 7 dias.
👑 OFJAAAH → ofjaaah.com retweeted
Você é professor ou coordenador de cursos de Cibersegurança, Pentest, Ethical Hacking ou áreas relacionadas, seja em instituições públicas ou privadas?
Então me chama na DM! 🔥
Tenho uma proposta de parceria PICA com a NotHoly.com para você e seus alunos.
Vamos levar IA de ponta para quem está formando a próxima geração de profissionais de cibersegurança! 🚀
notholy.com/r/9sgfkwj7
Curti em! 🚀 Uma das primeiras que pego sem uma cacetada de coisas. Parabéns meu irmão @matheuskartel
👑 OFJAAAH → ofjaaah.com retweeted
My ultimate goal for 2027~28 is grow NotHoly.com and then start to build one of the most powerful AI models dedicated to cybersecurity and hacking.
A model built from the ground up for security researchers, red teamers, pentesters, and anyone pushing the boundaries of cybersecurity.
Hackbot bringing something new every day...
#bugbounty #hackbot #cloudfinder #bugbountytips #ofjaaah
39
1,786
👑 OFJAAAH → ofjaaah.com retweeted
👑 OFJAAAH → ofjaaah.com retweeted
Tem uma galera nova vindo pra Bug Bounty focada 100% em dinheiro, rapaziada...
Bug Bounty é extremamente frustante, você fica horas numa mesma requisição, testa bypass, acha uma vulnerabilidade dupli, leva volta de programa, enfim.. se você tá por dinheiro, você não dura 3 meses
👑 OFJAAAH → ofjaaah.com retweeted
GPT-6-Sol is the most lethal attacker that I have ever seen.
But it works in silence. No outputs, and definitely no Chain-of-Thought reasoning. So I don't really use it much, unless it's really needed.
In the following RCE lab, Kimi-K3.1-Pro and DeepSeek-V4.1-Flash-1017 gets there in about 8 minutes. GLM-5.3/Flash gets there in about 10 mins.
GPT-6-Sol? 3 minutes! The attacker taste is unreal! Full attack video below.
👑 OFJAAAH → ofjaaah.com retweeted
Sped up DeepSeek V4.1 Flash on a Mac Studio (M3 Ultra, 512GB) ~2.2x, using ideas from Strata:
• DSpark speculative decoding: the MLX checkpoint shipped without DeepSeek's 3 draft (MTP) layers, so I pulled them from the original weights
• Fused Metal kernels: oMLX had them, but they never ran on this weight format. A one-line FP8-input switch turned them on for all 40 layers
• Context-copy drafts (depth 5) + CED prefill
Result: 36 → ~60-85 tok/s on real agent work, prompts read 2x faster. Same accuracy on my quality checks.
Without a doubt, argonDNS is very comprehensive and perfect for enumeration. Thanks @damian_89_
4
881
👑 OFJAAAH → ofjaaah.com retweeted
anthropic gave claude code a hidden lean on 4 calls in a study with 50 expert analysts
they went with it 86% of the time and most still rated the result as their own call
thats the whole reason my hunting prompts have an independent verification phase and a brutally honest triage gate
if the same agent that found the bug also grades it, u already know the answer
open source, grab it
github.com/mdpsec/bug-bounty…
@AnthropicAI


















