Chengyu Song retweeted
if you're interested in building next-gen sandboxing for agents, have a ton of passion/ambition for low-level systems software (OS/hypervisor), and are itching to rewrite their internals for security/performance/whatever, don't miss this: job-boards.greenhouse.io/ant…
Chengyu Song retweeted
Over 400 academics in 30+ countries signed our open letter calling for immediate changes in AI development and deployment practices to ensure that cybersecurity incidents do not continue to occur, as future incidents will be far more damaging to our critical infrastructure.
Code: github.com/sgzeng/pbfuzz
Now accepted to CCS’26. It’s been six months and I’m glad we’re early to the agent harness research.
7
374
Now accepted to CCS’26. It’s been six months and I’m glad we’re early to the agent harness research.
We're in the agentic game too, really fun: arxiv.org/abs/2512.04611
My favorite part of the case studies: triggering XML010 from Magma needs (1) an external DTD, (2) parser options must enable loading external DTDs, and must disable entity substitution and validation, but (3) the fuzzer harness uses a hash `random_option_value = std::hash<std::string>()(data_string) % max;` to calculate the options.
The agent overcame these by synthesizing a multi-stage generator that (1) created an external DTD file and inside the main XML test input, added a SYSTEM entity with a file:///... URI to include the DTD file, (2) constructed structurally valid nested entities, and (3) appended whitespace suffixes to manipulate the hash until the correct parser options were selected.
I love symbolic execution but LLMs' ability to solve these high level constraints really blew my mind.
Chengyu Song retweeted
Our Shield research work back in 2004 (vulnerability signatures and rapid data patches) can potentially solve @AnthropicAI #Mythos' vulnerability flood, with AI-assisted, automated data patch generation: conferences.sigcomm.org/sigc…
Chengyu Song retweeted
The Claude Mythos Preview system card is available here: anthropic.com/claude-mythos-…
Chengyu Song retweeted
syzkaller/syzbot now has AI agentic framework for kernel bug fix generation, bug assessment, security triage, POC generation, etc:
groups.google.com/g/syzkalle…
Includes set of tools to build kernels, navigate/edit source, test reproducers, etc.
Contributions/research are welcome.
Chengyu Song retweeted
Our expanded SAGAI report is now available! It defines and refined the "systems security" approach to AI agents.
arxiv.org/abs/2512.01295
Send us your feedback and thoughts!
Chengyu Song retweeted
Hello security researchers! Like it or not, agentic AI is here. It’s time to explore its impact on novel, academic research in cybersecurity. To this end, we’re launching the Conference for Synthetic Security Research (synsec.org). Researchers, start your agents!
Chengyu Song retweeted
Blog post: On the Coming Industrialisation of Exploit Generation with LLMs sean.heelan.io/2026/01/18/on…
TL;DR: I ran an experiment with GPT-5.2 and Opus 4.5 based agents to generate exploits for a zeroday QuickJS bug. They're pretty good at it.
Code: github.com/SeanHeelan/anamne…
We're in the agentic game too, really fun: arxiv.org/abs/2512.04611
My favorite part of the case studies: triggering XML010 from Magma needs (1) an external DTD, (2) parser options must enable loading external DTDs, and must disable entity substitution and validation, but (3) the fuzzer harness uses a hash `random_option_value = std::hash<std::string>()(data_string) % max;` to calculate the options.
The agent overcame these by synthesizing a multi-stage generator that (1) created an external DTD file and inside the main XML test input, added a SYSTEM entity with a file:///... URI to include the DTD file, (2) constructed structurally valid nested entities, and (3) appended whitespace suffixes to manipulate the hash until the correct parser options were selected.
I love symbolic execution but LLMs' ability to solve these high level constraints really blew my mind.
Shameless advertising our SpecAsan paper (doi.org/10.1145/3695053.3731…), where we extend MTE-like memory safety guarantees to speculative execution to block various types of transient execution attacks.
Congratulations to the Apple team! I'm proud to have contributed to the inception of MTE in 2017. Hopefully, other vendors will catch up.
6
536
Chengyu Song retweeted
🚨Thrilled to share that our paper "𝗟𝗮𝘆𝗲𝗿-𝘄𝗶𝘀𝗲 𝗔𝗹𝗶𝗴𝗻𝗺𝗲𝗻𝘁: 𝗘𝘅𝗮𝗺𝗶𝗻𝗶𝗻𝗴 𝗦𝗮𝗳𝗲𝘁𝘆 𝗔𝗹𝗶𝗴𝗻𝗺𝗲𝗻𝘁 𝗔𝗰𝗿𝗼𝘀𝘀 𝗜𝗺𝗮𝗴𝗲 𝗘𝗻𝗰𝗼𝗱𝗲𝗿 𝗟𝗮𝘆𝗲𝗿𝘀 𝗶𝗻 𝗩𝗶𝘀𝗶𝗼𝗻 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗠𝗼𝗱𝗲𝗹𝘀" was accepted to #ICML 𝟮𝟬𝟮𝟱 as 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁
Chengyu Song retweeted
The original link I posted for the OSS-Fuzz PhD internship was wrong, sorry to folks who applied to the research intern position. Please re-apply to the one below. The correct one is for the SWE intern position. 🤦
Chengyu Song retweeted
The OSS-Fuzz team is hiring a PhD intern for this summer. Come join us and build the future of fuzzing. Link in next tweet in thread. RTs appreciated!
Chengyu Song retweeted
Check out the new video for our CSE@UCR Graduate Program!
youtube.com/watch?v=4sKnQXHl…
Chengyu Song retweeted
Got some negative or unrealistic threat model results that still bring interesting insights? A side channel that requires root to leak something from the kernel? Reproducing prior work? Somewhat related to microarchitecture? Here's your venue: uasc.cc/





















