Anthropic’s New Claude Code Security Feature Can Now Scan Codebases and Suggest Targeted Fixes

Anthropic, the company behind Claude, has introduced a new feature called Claude Code Security, which is designed to help organisations detect and fix security vulnerabilities in their software. The tool is built into Claude Code on the web and is currently available in a limited research preview for Enterprise and Team customers. It is also being offered to open-source maintainers through special access. What it does is use AI to scan codebases, identify security weaknesses, and suggest software patches for human review.

Here is more to it.

Claude Code Security – A New Feature to Find Vulnerabilities

With a growing number of software vulnerabilities and limited resources to address them, security teams often face trouble. Looking at traditional security tools, which mainly rely on static analysis, they scan code for known vulnerability patterns like exposed credentials or outdated encryption methods, but frequently miss more subtle and complex issues.

On the other hand, Claude Code Security reads and analyses code in a way similar to a human security researcher. Further, it examines how different components interact, traces how data flows through applications, and identifies deeper issues such as business logic flaws or broken access controls.

It uses a multi-stage verification process to reduce false positives and re-checks its own findings. It attempts to confirm or disprove them while also assigning severity ratings. Not only this, but it also assigns a confidence rating to each issue it finds so that developers can better understand the likelihood and impact of the vulnerability.

Although it suggests security patches, the final implementation of the patch has to be done by a human reviewer. This way, human experts remain in control of the final decisions.

As per Anthropic, it took nearly a year of cybersecurity research to build this tool. Adding more to this, the company also claims that by using Claude Opus 4.6, the internal team was able to find over 500 vulnerabilities in production open-source codebases, including issues that had gone undetected for years.

While attackers may use AI to find weaknesses more quickly, defenders now have this Claude Code Security tool, which can also use similar technology to secure systems faster and more effectively.

And as stated above, the research preview is now open to Enterprise and Team customers, and open-source maintainers can apply for expedited access. The company also states that after refining the tool through further feedback, it plans to go for a wider release as well.