cenzhang

[Satya Nadella @satyanadella]

Our new multi-model agentic security system brings together more than 100 specialized agents across frontier and custom models to find exploitable bugs, delivering top performance on the CyberGym benchmark.

We used it ahead of Patch Tuesday to help find and fix 16 vulnerabilities. Today we’re announcing that customers can sign up to test it in private preview.

https://t.co/maAN55yZQ1

[cenzhang @zhclhy]

🔹 Ensemble wins: generate 3, pick the best.
🔹 All open-sourced in OSS-CRS: https://t.co/WCvVQP1qBh
#CyberSecurity #AI #LLM #Patch
(3/3)🧵

[cenzhang @zhclhy]

How good are AI coding agents at patching zero-day vulnerabilities in 2026?
Our team @GeorgiaTech tested 10 agent configs (4 frameworks × 5 frontier models x 1k+ manual patch reviews) on DARPA AIxCC vulnerabilities.
Full post 👇 :
https://t.co/60pbgsS3Sr
(1/3) 🧵

[cenzhang @zhclhy]

Key takeaways:
🔹 Claude 3.7 Sonnet 2025 (~52%) → Copilot + GPT-5.3 2026 (~71%).
🔹 Model diff > framework diff. Native frameworks show no clear advantage.
🔹 20–40% semantic incorrectness gap: passing automated tests ≠ correct patch.
(2/3) 🧵

[Team Atlanta @TeamAtlanta24]

🔎 How do sink-aware fuzzing + LLMs uncover Java vulns under pressure? See how Atlantis-Java did it.

👉 https://t.co/VhWicuIaVq

#AIxCC #Cybersecurity #CTF #AI #LLM #GenAI #AICyberChallenge #DARPA #DEFCON #Security #Vulnerability

[cenzhang @zhclhy]

We are the best!

[cenzhang @zhclhy]

🏆Excited to share that our paper "Semantic-Enhanced Static Vulnerability Detection in Baseband Firmware" has won ACM SIGSOFT Distinguished Paper Award at #icse2024! Congrats to Yiming @AlexMyFavorite , myself, and other co-authors! Special thanks to my supervisor Prof. Liu Yang @Sun450gmailcom1 and my wife!
Paper and prototype are open: https://t.co/XPMKLeBSMA, https://t.co/jlC3uGmpYo

[cenzhang @zhclhy]

@palinatolmach @NTUsg @rv_inc @summit_defi congrats😁