Research
2026
SuperRed: AI Red-Teaming and Security Benchmarking with Fine-grained Threat Model ICLR Video
Simon Sure, Sarthak Munshi, Rishabh Sinha, Zhun Wang, Dawn Song
Defense Trilemma: Why Prompt Injection Defense Wrappers Fail? Preprint
Manish Bhatt, Sarthak Munshi, Vineeth Sai Narajala, Idan Habler, Ammar Al-Kahfah, Ken Huang, Joel Webb, Blake Gatto, Md Tamjidul Hoque
Manifold of Failure: Behavioral Attraction Basins in Language Models Preprint
Sarthak Munshi, Manish Bhatt, Vineeth Sai Narajala, Idan Habler, Ammar Al-Kahfah, Ken Huang, Blake Gatto
2025
MAPLE: Specialized Modular Neural Architectures for Multi-Agent RL PDF
Zahra Sheikhbahaee, Chang Su, Sarthak Munshi, Aniket Didolkar, Juan David Vargas, Adam Safron, Nicco Reggente, Leonardo Christov-Moore, Dianbo Liu, Irina Rish
Representation Control for LLMs: Survey and Research Challenges ACM Computing Surveys Preprint
Lukasz Bartoszcze, Sarthak Munshi, Bryan Sukidi, Jennifer Yen, Zejia Yang, David Williams-King, Linh Le, Kosi Asuzu, Carsten Maple
ACSE-Eval: Can LLMs threat-model real-world cloud infrastructure? ESORICS Workshops Preprint
Sarthak Munshi, Swapnil Pathak, Sonam Ghatode, Thenuga Priyadarshini, Dhivya Chandramouleeswaran, Ashutosh Rana
Code
01
SuperRed – AI red-teaming and security framework.
02
TurboKV – High-performance key-value store.
03
UnSQL – Natural language to SQL for legacy databases. ACQUIRED
04
Chapel Crypto – Cryptography primitives for Chapel's official package ecosystem.
Public vulnerability disclosures / CVEs
Heap overflow-based RCE in GGUF loading // apple mlx
Stack overflow in SSPI auth // squid proxy
Heap overflow in cache digest // squid proxy
Timing side channel in sasl server // memcached
Format string buffer overflow in avformat // ffmpeg
Authorization bypass in S3 bucket consumption // apache mlflow