Claude Security enters public beta

Anthropic launches Claude Security in public beta for Enterprise users, using Claude Opus 4.7 to scan GitHub repositories for software vulnerabilities.

Anthropic has launched Claude Security in public beta for its Enterprise customers, following a private preview deployed under the name Claude Code Security with several hundred organizations. The tool scans codebases for vulnerabilities and offers targeted fixes, without requiring API integration or a custom agent.

The service is built on Claude Opus 4.7 and can be accessed from the sidebar of claude.ai or at claude.ai/security. The user selects a GitHub repository, a folder, or a branch, then initiates a scan. The model is not limited to searching for known patterns: it traces data flows between files, examines interactions between components, and reads source code in context, much like a security researcher. Each identified vulnerability is accompanied by a confidence score, a severity assessment, reproduction steps, and a patch instruction that can be opened directly in Claude Code on the Web.

This public release introduces scheduled scans, sub-directory targeting, CSV or Markdown export, sending results via webhook to Slack or Jira, and the ability to triage a finding by logging the reason for rejection for future reviews. A multi-stage validation process filters false positives before they are escalated to the analyst.

In parallel, Opus 4.7's capabilities are integrated into third-party platforms via technology partners CrowdStrike, Microsoft Security, Palo Alto Networks, SentinelOne, TrendAI, and Wiz, as well as into the systems of service partners Accenture, BCG, Deloitte, Infosys, and PwC. Access for Team and Max plans is scheduled for a later stage. This initiative extends Project Glasswing, a program through which Anthropic makes its Mythos Preview version available to a restricted circle of partners for the detection and exploitation of software vulnerabilities.