Fuzzilli vs. BigSleep: How LLM‑Based Bug Finding Complements Traditional Fuzzing in Chrome’s V8 Engine
What Happened — Google Project Zero evaluated two automated vulnerability‑discovery tools that run continuously against Chrome’s JavaScript engine, V8. The generative fuzzer Fuzzilli reported 37 bugs in the past six months, while the LLM‑driven system BigSleep reported 45, with only 25 % of the findings overlapping. Separate overlap experiments suggest the true overlap may be around 60 %. The study shows that each technique uncovers a distinct set of flaws, confirming that fuzzing and LLM‑based analysis are complementary rather than interchangeable. Project Zero recommends that vendors adopt both approaches to maximise pre‑release vulnerability coverage.
Why It Matters for Trust & Control Assurance — The findings highlight the need for a control objective that ensures multiple, independent vulnerability‑detection methods are in place, providing stronger evidence for audit readiness.
Who Is Affected — Browser vendors, cloud‑service providers, and any organization that ships software containing complex runtime engines (e.g., JavaScript, WebAssembly).
Recommended Actions – Add an independent LLM‑based analysis pipeline alongside existing fuzzing suites; map both to your vulnerability‑management controls and retain evidence of coverage for auditors. Source: https://projectzero.google/2026/10/fuzzing-v-AI.html
Technical Notes – Fuzzilli is a generative JavaScript fuzzer; BigSleep is an LLM that reviews new commits, writes PoC test cases, and submits bugs via Clusterfuzz. No specific CVEs are disclosed; the study focuses on detection methodology. Source: https://projectzero.google/2026/10/fuzzing-v-AI.html