Anthropic Announces Claude Security

Ronni Holmvig Strøm · 2026-02-21

Anthropic just dropped a major update in the security space with Claude Code Security, a new built-in capability for Claude Code on the web.

Announced February 20, 2026, it's currently in limited research preview for Enterprise and Team customers (with expedited free access available for open-source maintainers via application).

Key Highlights

Core Functionality Scans entire codebases for security vulnerabilities (including subtle logic flaws, business-logic issues, and broken access controls that rule-based tools miss), then suggests targeted patches for human review and approval.

Powered by Opus 4.6 Leverages the recently released Claude Opus 4.6 model, which Anthropic says is "notably better" at high-severity vulnerability detection. In internal testing, it uncovered over 500 previously unknown vulnerabilities in production open-source codebases—some bugs that had survived expert reviews and fuzzing for decades.

Defensive Focus Anthropic emphasizes making frontier defensive capabilities widely available to level the playing field for cyber defenders. They already use Claude internally to review and secure their own systems.

Development Background Refined over a year via internal red-teaming, Capture the Flag competitions, and collaboration with Pacific Northwest National Laboratory for accuracy.

Market Impact

The announcement triggered immediate turbulence in cybersecurity stocks:

CrowdStrike, Cloudflare, Palo Alto Networks, Zscaler, Okta, JFrog, and others saw sharp declines (e.g., -8% to -25% in some cases), wiping out billions in market value within hours.

Analysts describe it as another "AI disruption" wave hitting legacy security incumbents, similar to recent SaaS and financial-data sell-offs.

This positions Anthropic (alongside moves from OpenAI and Google) as a direct player in AI-driven DevSecOps and vulnerability management—shifting from pure model releases to applied security tooling.