Benchmark report
Why do you need us for AI agent security?
SolonGate, Claude Code permissions, Invariant, llm-guard and roll-your-own guardrails, measured on attacks blocked and legitimate work broken.
- 1,652 harmful tool calls
- 24,911 real tool calls
Measured, with the method attached.
Technical reports by SolonGate. They are not peer-reviewed, and each one evaluates SolonGate’s own products alongside others.
All publications
Benchmark report
SolonGate: measuring the scanner behind Shadow AI
SolonGate’s scanner against Microsoft Presidio, gitleaks and trufflehog — including the rows where it loses and the bugs the benchmark found in it.
- 1,890 real pastes
- 4,930 marked values
- 1.1M tokens of clean code
Publication formats
- Benchmark report2
- Technical report0
- Research note0
- Dataset0
Found an error or a failed reproduction? We want to know.
Send a correction