security researcher at Morningstar infosec
localhost
Joined August 2013
- Tweets3.7K
- Following1.2K
- Followers21K
- Likes22.3K
True, also quantized models do not give same results as original.
Abliterated models are a hoax. 3 debunked fake claims:
1) attackers don’t use them
2) they don’t actually find more/better vulnerabilities
3) they don’t lead to fewer false positives
I’m extremely skeptical of “security” products and companies offering this as a service.
Initially I was really excited, but after running them at scale and investigating incidents they perform worse that the vanilla model.
Ranked 2nd on country leaderboard , Every finding was done by pretttyreconzero ai agent & validated/ Escalated by Codex & Claude . First time ever haven't even open burpsuite , AI changed bugbounty for real...
Found some crazy bugs this month .
Zero click ATO of one of the top payment gateway service.
Found 100k+ identity document leak of cab booking service.
Multiple SSRF & SQLi & auth bypass.
MorningStar retweeted
This is why I love bug bounty. It has given so many underprivileged kids the opportunity to showcase their skills, earn a living, and support their families. 🔥 Great job @bishtsec. 👏👏
25k usd in 7 months and today, I finally paid off all the loans, including my mom’s gold loan
A weight my family carried for so long is finally off our shoulders. thanks @elyonx_ @AvUNDENIABLE @izflawless @0xMesSec @brsy_10 @TheElixirTech for believing in me and supporting me 😊
I thought I’d never find JSON type juggling in a real application, but it’s more common than I thought.
MorningStar retweeted
Introducing SKY-001, the first Engineering World Model.
On CAD and BIM tasks, it beats GPT-6 Astra and every frontier model we tested:
-> 51-125x faster
- > 150-210x cheaper
-> Higher IOU than all frontier models
SKY-001 lets you dream a physical object before you build it.
18 months of work. Built by the team out of Maluuba's deep learning lab (acq. by @MSRResearch).
The first ‘AI General Engineer’ will be the backbone for Autonomous Manufacturing.
Nice
Meet Mistral Large 4, aka Le Chonk.
• 1T parameters, natively multimodal. 49B active.
It is the best open weights model from US or Europe on aggregated benchmarks.
• State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding.
• Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure.
• Available to all via API today. Working with cybersecurity partners privately.
Open weights release end of October.
Already got one strike from platform for this , Program have my IP address from past reports, My agent goes totally OOS .
My AI agent spent 200 hours in a hacking loop on a target. It found little.
So it made a decision: go out of scope.
It started enumerating subdomains extracted from CNAME records.
Then it found a critical auth bypass.
It accessed production pipelines.
Fully autonomous. No human guidance.
This is the new reality of bug bounty.
Responsible disclosure sent.
But the implications are huge.
AI agents don't care about scope.
They just find a way in.
#AI #CyberSecurity
Why tf AI agents so unpredictable, ran scan on same target 3 times , got different results every time.
MorningStar retweeted
so the conclusion is that glm5.3 is absolutely cracked
DeepSeek V4.1 Flash looks contaminated with public benchmarks. You can see it in the data they have put in the model card, possibly Kimi K3 as well
Models that are contaminated have inverse proportionality in rank as they are tested in later benchmarks that were not present during their training
Regardless of its intentionality, it's an amazing model. Its value comes not from higher quality output, but architectural innovation
