AI Security Measurement

Static analysis can find real defects in an agent's software substrate, but a clean report cannot prove that model-mediated intent, framework semantics, delegated authority, and multi-step behavior are secure. This series measures that boundary and develops an honest assurance envelope for reporting what was checked, what was missed, and what remains unknown.

1 series · 1 article

Static-Analysis Confidence Gap

1What a Clean Static-Analysis Report Actually Proves

2Measuring Static Analysis on AI-Agent Vulnerabilitiesplanned

3The Agent-Framework Modeling Gapplanned

4Model-Mediated Taint Trackingplanned

5Modeling Agent-Specific Sources, Sinks, and Trust Boundariesplanned

6False Positives and Function-Breaking Remediationplanned

7Agentic Risks Beyond Source-to-Sink Analysisplanned

8When Safe Operations Compose into an Unsafe Planplanned

9Static Analysis as a Gate for AI-Generated Codeplanned

10An Honest Assurance Report for Agent Softwareplanned