We Scanned 16 AI Agent Repos. 76% of Tool Calls Had Zero Guards.

I scanned 16 open-source AI agent repos with AST analysis. 76% of tool calls that write to databases, send payments, or delete files have zero security checks.