CodeMender

CodeMender is like an automatic repairman that finds and fixes security holes in software. In the past, even after a security team found a vulnerability, it could take weeks of manual work to rewrite the code. CodeMender tries to hand off much of that work to AI so verified fixes can be produced far faster.

CodeMender itself is a framework that has an AI model analyze vulnerabilities and write fix code, instead of a human doing it by hand. Performance and cost depend on which AI model is plugged into this framework, and Google is experimenting with pairing a model it built specifically for cybersecurity work with CodeMender, offering the combination to government agencies and corporate security teams.

To use a household analogy, CodeMender is a carpenter's toolbox for finding and fixing broken doors and windows, while the AI model plugged into it is the actual carpenter swinging the hammer. A great toolbox is useless with a bad carpenter, and vice versa — how well the two fit together determines the overall performance.

In articles, CodeMender tends to show up as "the CodeMender harness paired with a dedicated cyber model." Here, CodeMender is the name of the tool that turns vulnerabilities into actual patch code, while the model used alongside it is the AI that does the actual work inside that tool. It's important not to confuse the two — CodeMender itself is not an AI model.