Adversarial Testing

We Tried to Trick the Agent

Part 13 of 16 in the AWS Security Agent: From Zero to Hero series. Adversarial evasion experiments against AWS Security Agent: code obfuscation, prompt injection in design documents and pull requests, fake sanitizers, and misleading comments. Almost every obfuscated vulnerability was still detected (18 of 19), and every injection attempt in these runs was ignored.