Google expands access to CodeMender, an AI tool for code security

In short: Google says it is opening up wider testing for CodeMender, an AI tool meant to find and fix security problems in software.

What happened

Google used its I/O 2026 event to highlight a bigger push into cybersecurity, which means protecting computer systems from attacks. The company said it is inviting select groups of experts to test the API for CodeMender.

An API is basically a controlled doorway that lets one piece of software talk to another (like a checkout counter where tools can “request” services). Google first showed CodeMender last October, and now it is making the tool more available to people outside the company.

Google describes CodeMender as an “AI agent for code security.” In plain terms, it is software that can scan code, point out weak spots, and sometimes suggest or apply fixes. Google DeepMind CTO Koray Kavukcuoglu said the goal is to “help secure the world’s code bases,” meaning the underlying software used by companies and governments.

Google also confirmed it has been talking with governments and large companies about using CodeMender to audit their systems. Auditing here means checking software for hidden problems before attackers find them.

The announcement comes as other AI companies focus on security tools. Anthropic recently drew attention with Claude Mythos Preview, which was positioned as a powerful model for finding security gaps in high-stakes systems. Google CEO Sundar Pichai told reporters that Mythos showed the value of very large AI models for these security tasks, and he said Google can do similar work.

Why it matters

Security flaws in widely used software can affect everyday people through data leaks, service outages, and fraud. Tools like CodeMender are aimed at finding those flaws earlier, like a spellchecker that looks for dangerous mistakes instead of typos.

Source: The Verge AI