Fuzzers vs. AI: Comparing Fuzzilli and BigSleep on JavaScript Engines
Web page · https://projectzero.google/2026/10/fuzzing-v-AI.html
First seen by Oriel on .
AI summary
Project Zero finds that BigSleep reported slightly more actionable V8 vulnerabilities than Fuzzilli over six months, but the tools discover many different bugs and remain complementary.
- From April to September 2026, BigSleep found 45 actionable vulnerabilities and Fuzzilli found 37; their reported bugs overlapped by roughly 25%.
- A historical rediscovery experiment had BigSleep recover about 62% of 112 Fuzzilli vulnerabilities, find a different fixed bug for another 12%, and find none for 26%.
- The experiment used Gemini 3.6 flash for 25 attempts per commit, with crashes checked against pre-fix and post-fix revisions.
- The authors found no statistically significant evidence of memorization before 2026, but recommend reevaluating this for other models and future generations.
- Harder-to-fuzz bugs also tended to be harder for the agent to find, and overlap was notably smaller for compiler bugs. Cost per bug was not yet comparable.
#v8 #fuzzing #llm agents #vulnerability research
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Points
Score reported by the source.
- Lobsters: 18, latest observation ; +15 over observed period since
Comments
Comment count reported by the source.
- Lobsters: 4, latest observation ; +4 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Lobsters | rswyzt | https://projectzero.google/2026/10/fuzzing-v-AI.html Read discussion |