
CodeMender is an AI code security agent in Agent Platform that finds, verifies, and fixes vulnerabilities in application source code across languages including Java. Currently in Public Preview for a limited set of customers, it works directly on a service's repository rather than screening prompts and responses, detecting sensitive data, or governing agent traffic. Keeping that boundary straight matters for the Professional Agentic Architect exam, because several other security services sit alongside it without ever inspecting or repairing source code.
When run against an application repository, CodeMender works through finding, verifying, and fixing vulnerabilities in sequence. To confirm a finding, it builds the code and attempts an exploit. Confirming the weakness first establishes that the vulnerability is real before a fix is produced.
After verifying the issue, CodeMender generates a patch and tests it. Human confirmation of its actions is on by default, so engineers review the proposed patch before it is applied. With human confirmation enabled by default, engineering teams stay in control of which changes reach the repository.
Google Cloud's AI security services each work on a different layer, and placing CodeMender among them comes down to where a security issue lives:
Matching each service to its layer keeps the choices clear. Content screening belongs to Model Armor, sensitive data detection to Sensitive Data Protection, agent traffic to Agent Gateway, and vulnerabilities in application code to CodeMender in Agent Platform.
Our Professional Agentic Architect practice exams test CodeMender alongside Model Armor and Agent Gateway, with detailed explanations that drill these distinctions.