Understand this story
Karl Weinmeister, Director of Developer Relations at Google, introduced a process for finding Go bounds-check defects: first, use the runtime.panicBounds traps remaining after Go compiler SSA optimization to locate candidate crash points; then, Gemini 4 Argon translates the Go source code and RISC-V disassembly into Z3 constraints; after Z3 derives an input that can trigger the issue, it is handed to go test for reproduction. This process hit the target on all three parsers with implanted bugs, and also found three potential bugs in skill-lint. Karl Weinmeister also pointed out that go test -fuzz misses out-of-bounds crashes because slices often carry excess capacity; after trimming slice capacity to length, 3 seconds of fuzzing was enough to reproduce two out-of-bounds bugs.
AI-generated from public reports
Latest development · Report
Using Gemini and the Z3 constraint solver to find bugs that Go fuzzing missesDevelopments
1 confirmed developments
Karl Weinmeister introduces the workflow of using Gemini and Z3 to find Go bounds-check defects
Source:Google Developers Blog
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 12
Google Developers BlogFirst-party
Using Gemini and the Z3 constraint solver to find bugs that Go fuzzing misses
Google Director of Developer Relations Karl Weinmeister introduces a workflow: first use the runtime.panicBounds traps left after Go compiler SSA optimization to locate candidate crash points, then have Gemini 4 Argon translate the Go source code and RISC-V disassembly into Z3 constraints, and after Z3 derives a triggering input, hand it to go test for reproduction. This workflow hit every time on three parsers with implanted bugs, and also found three potential bugs in skill-lint. The author also points out that go test -fuzz misses out-of-bounds crashes because slices often carry excess capacity; after trimming slice capacity to length, a 3-second fuzz run reproduces two out-of-bounds bugs.