We blame the bot when AI-generated code is vulnerable, but what if it is just using the dangerous APIs we left on the table?

This talk tests whether API design can steer AI coding agents toward safer code. The same agent builds the same Java/Spring task app twice: once in a normal environment with freedom to choose libraries and APIs, and once in a constrained one with secure-by-default scaffolding, deny-by-default settings, safer abstractions, and bans on risky APIs such as raw path handling.