Anthropic launched beta testing of Claude Security—a service based on Opus 4.7 for detecting vulnerabilities in code

Anthropic launches open beta testing of Claude Security

The company Anthropic announced the start of public testing for its new product—Claude Security, which is intended for clients of the enterprise version of Claude Enterprise. Previously it was called Claude Code Security and runs on the Claude Opus 4.7 model.

What does Claude Security do?
* Scans source code to identify vulnerabilities.
* Provides targeted fixes directly within the project context.
* Does not rely solely on templates; it analyzes component interactions, data flows, and real threats—just like a cybersecurity specialist would.

After detecting each issue, the service issues:

1. An explanation with a confidence rating.
2. Severity level and potential impact.
3. Steps to reproduce the vulnerability.
4. Specific remediation recommendations.

The user can open the identified problem in Claude Code on the Web—the browser version of Claude Code—to apply a patch immediately without switching to an IDE or terminal.

How to gain access
* Via the sidebar on the Claude.ai website.
* Or directly at claude.ai/security. The user selects a repository, optionally limits the scan to a specific directory or branch, and starts scanning.

Why it matters
Anthropic emphasizes that AI significantly shortens the interval between vulnerability discovery and exploitation. New models can already not only find gaps but also automatically exploit them. Recently the company introduced Claude Mythos Preview—a model that, according to developers, finds and exploits vulnerabilities at a level comparable to leading cybersecurity experts. Mythos was available to a limited group of partners under the Project Glasswing initiative, while Claude Security is aimed at a broader range of corporate clients.

Testing and effectiveness
* Over two months of testing, the product passed validation in hundreds of organizations.
* A multi-stage pipeline automatically validates each finding before it reaches analysts, assigning it a confidence score. Analysts receive only signals that require attention.
* Several teams have already completed the journey from scanning to patch application in a single session, instead of days of back‑and‑forth between security and engineering.
* The ability to schedule regular checks allows replacing one‑off audits with continuous monitoring.

Partners and integrations
Technology partners:
CrowdStrike, Microsoft Security, Palo Alto Networks, SentinelOne, TrendAI, Wiz—these companies are integrating Opus 4.7 capabilities into their existing security platforms.

Service partners:
Accenture, BCG, Deloitte, Infosys, and PwC, together with corporate security services, are deploying Claude‑based solutions for vulnerability management, code analysis, and incident response.

Availability
Users on the Claude Team and Claude Max plans will gain access to the new service in the near future.