Back in February, Anthropic launched Claude Code Security as a “Limited Research Preview” — an AI tool that scans codebases for vulnerabilities and suggests patches. Now, two months later, Anthropic is opening the doors wider: Claude Security moves into public beta for Enterprise and Team customers.
What’s changed
The biggest difference from the preview: five major cybersecurity companies are now integrating Claude Opus 4.7 directly into their platforms. CrowdStrike, Palo Alto Networks, SentinelOne, TrendAI (Trend Micro’s AI division), and Wiz are bringing Anthropic’s vulnerability detection into the tools enterprises already use. That’s a clear signal — Anthropic isn’t treating security as a niche product. It’s building it as an infrastructure layer.
There’s also a new ability to schedule scans. Security teams can now set up a regular cadence instead of manually triggering reviews.
How Claude Code Security works
The approach is fundamentally different from traditional static analysis tools. Instead of pattern-matching against known vulnerabilities, Claude reads code the way a human security researcher would: tracing data flows, understanding how components interact, and catching complex business logic flaws that rule-based scanners miss.
Every finding goes through a multi-stage verification process. Claude actively tries to disprove its own findings before presenting them to an analyst. Each finding comes with a confidence rating — nothing gets applied without human approval.
The numbers speak for themselves
Using Claude Opus 4.6, Anthropic’s Frontier Red Team found over 500 vulnerabilities in production open-source codebases — bugs that had gone undetected for decades despite years of expert review. With Opus 4.7, those capabilities are likely even stronger.
My take
Cybersecurity is the strongest enterprise use case Anthropic has found so far. No other domain so directly connects “AI finds what humans miss” with “and it costs serious money if you don’t find it.” The fact that five of the biggest security firms are integrating Opus 4.7 tells you this isn’t an experiment anymore.
Sources: