Our paper, "Fuzzilicon: A Post-Silicon Microcode-Guided x86 CPU Fuzzer," received a Distinguished Paper Award at NDSS Symposium 2026, one of the top venues in network and systems security. The work is co-authored by Sentinel Intelligence co-founder Lichao Wu, together with Johannes Lenzen, Mohamadreza Rostami, and Ahmad-Reza Sadeghi (Technical University of Darmstadt).
What Fuzzilicon does
Fuzzing feeds a system huge numbers of generated inputs and watches for unexpected behavior. For CPUs, that has mostly meant fuzzing a design in simulation, which needs access to the chip's source design and can miss what happens on real silicon. Fuzzilicon is the first post-silicon fuzzing framework for real-world x86 CPUs. It uses reverse-engineered access to Intel's microcode update interface to instrument the microcode, getting feedback from inside the processor while it runs.
- Works on the finished chip, with no access to its design files and no vendor support.
- Applied to Intel's Goldmont microarchitecture, it reported five findings, including two previously unknown microcode-level speculative-execution vulnerabilities.
- It automatically rediscovered the µSpectre class of vulnerabilities, which earlier work had found by hand.
- Up to 31× lower coverage-collection overhead than baseline techniques.
The rediscovery result is the one worth sitting with. A class of bugs that once took manual expert analysis to find now falls out of an automated, coverage-guided search.
Our take
The systems we depend on most are often ones we can't see inside, whether that's a processor or a commercial AI model. Fuzzilicon shows that rigorous, automated testing is still possible from the outside, without the vendor's cooperation. That's the same principle behind independent verification of AI agents: you shouldn't have to take a system's safety on trust.