On February 20, 2026, Anthropic rolled out Claude Code Security, a new capability in the Claude Code web experience, as a limited research preview. The feature scans codebases for security vulnerabilities and proposes targeted patches for human review, aiming to help defenders find and fix issues that traditional tools often miss.
Why this matters
Security teams commonly face large backlogs of vulnerabilities and limited staffing. Traditional static-analysis tools are typically rule-based and look for known patterns, which works well for obvious problems such as exposed credentials or outdated encryption. However, more complex, context-dependent flaws — for example business-logic errors or broken access controls — frequently evade those tools and require skilled human researchers to detect.
Claude Code Security takes a different approach: instead of matching known patterns it reads and reasons about code more like a human security researcher. It models component interactions, traces how data flows through an application, and can surface complex vulnerabilities that rule-based scanners miss.
Verification, prioritization and human oversight
Every finding goes through a multi-stage verification pipeline before being presented to an analyst. Claude re-examines its own results to try to prove or disprove them, reducing false positives. Findings are assigned severity ratings so teams can prioritize the most important fixes first, and each result includes a confidence score because some vulnerabilities are difficult to judge from source code alone.
Validated findings appear in the Claude Code Security dashboard, where teams can inspect suggested patches and approve changes. Nothing is applied automatically: Claude identifies problems and suggests fixes, but developers retain final approval.
Background research and results
Claude Code Security builds on over a year of Anthropic’s cybersecurity research. The company’s Frontier Red Team has been stress-testing Claude’s capabilities: running it in competitive Capture-the-Flag events, collaborating with Pacific Northwest National Laboratory on experiments to use AI to defend critical infrastructure, and iterating on Claude’s ability to find and patch real vulnerabilities.
As a result of this work, using Claude Opus 4.6 (released earlier in the month), Anthropic’s team found over 500 vulnerabilities in production open-source codebases — bugs that had gone undetected for decades despite prior expert review. Anthropic is currently handling triage and responsible disclosure with maintainers and plans to expand security collaboration with the open-source community.
Anthropic also uses Claude to review its own code and reports the model has been highly effective at improving the company’s internal security. Claude Code Security is intended to make those defensive capabilities available more broadly; because it’s built into Claude Code, teams can review findings and iterate on fixes within tools they already use.
The road ahead
Anthropic calls this a pivotal moment for cybersecurity: they expect a significant portion of the world’s code will soon be scanned by AI, given the models’ ability to find long-hidden bugs. That same capability could help attackers find exploitable weaknesses faster, so Anthropic argues defenders must move quickly to locate and patch issues to reduce risk. Claude Code Security is presented as one step toward higher security baselines across the industry.
Getting started
Claude Code Security is opening in a limited research preview today to Enterprise and Team customers. Participants will receive early access and work directly with Anthropic to refine the tool. Open-source maintainers are encouraged to apply for free expedited access.
For more information, Anthropic points interested parties to claude.com/solutions/claude-code-security.
