Anthropic Launches Claude Security for Enterprise Codebase Vulnerability Scanning

Anthropic has opened Claude Security to public beta for enterprise customers, deploying its most capable production model — Claude Opus 4.7 — as a full-spectrum vulnerability scanner that reasons about code the way a human security researcher would, not the way a regex engine does.

The product, which became available to all Claude Enterprise customers in early May, can be accessed directly from the Claude.ai sidebar or at claude.ai/security. No API integration or custom agent build is required — a deliberate design choice that signals Anthropic is pitching this as an out-of-the-box security tool rather than a developer primitive.

Beyond Pattern Matching

The distinction Anthropic is drawing is between signature-based detection and architectural reasoning. Traditional static analysis tools maintain libraries of known vulnerability patterns and flag code that matches them. Claude Security, by contrast, traces data flows, reads source code, and examines how components interact across files and modules — attempting to model what a skilled attacker would actually exploit.

That means the tool can surface vulnerabilities that have no prior signature: novel privilege escalation paths, authentication bypass patterns that emerge from the interaction of two individually safe components, and architectural weaknesses that only become dangerous under specific runtime conditions. Researchers testing Opus 4.7 in related contexts have reported the model independently identifying authentication bypasses in web applications, weaknesses in widely used cryptography libraries covering TLS, AES-GCM, and SSH, and a guest-to-host memory corruption flaw in a production virtual machine monitor.

A multi-stage validation pipeline independently examines each finding before it reaches an analyst, which Anthropic says drives down false positive rates — historically the bane of enterprise security tooling. Security teams that have spent years training analysts to ignore noisy scanner output will be watching that claim closely.

The Glasswing Backstory

Claude Security does not exist in isolation. It is the commercial, enterprise-accessible face of a broader initiative Anthropic announced in April 2026 called Project Glasswing — a cybersecurity consortium that brings together Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks as launch partners. More than 40 additional organizations that build or maintain critical software infrastructure have been granted access.

The centerpiece of Glasswing is Claude Mythos Preview, an unreleased frontier model that Anthropic describes as capable of surpassing all but the most skilled humans at finding and exploiting software vulnerabilities. Mythos Preview has already found thousands of high-severity vulnerabilities, including findings across every major operating system and web browser. Anthropic has committed up to $100 million in usage credits and $4 million in donations to open-source security organizations to support the initiative.

Claude Security, running on Opus 4.7, is the production-ready version of that capability — the model enterprises can deploy today while Mythos Preview remains in controlled research access.

Palo Alto Networks, one of the Glasswing launch partners, is among the security vendors integrating Opus 4.7 capabilities into their own platforms. CrowdStrike, Microsoft Security, SentinelOne, TrendAI, and Wiz are doing the same — a sign that Anthropic is pursuing a platform play rather than trying to win the endpoint security market outright.

KPMG Partnership Adds Audit Dimension

Separately, Anthropic and KPMG announced a global alliance on May 19 that will bring Claude to KPMG's 276,000-person workforce through the firm's Digital Gateway platform. Cybersecurity, risk, and AI assurance are named as core areas of the deployment — meaning KPMG professionals will use Claude to find and fix vulnerabilities and protect critical systems for the firm's clients, which include major private equity firms and tax clients worldwide.

The KPMG angle matters because it brings a professional services dimension to what might otherwise read as a pure technology story. Big Four firms do not adopt new AI tooling casually; their client engagements are governed by liability frameworks, audit standards, and regulatory expectations that require defensible, reproducible results. KPMG embedding Claude in a security context signals that at least one major professional services firm believes the output is reliable enough to put in front of regulated clients.

The Competitive Landscape

Anthropic is entering a market with established players. GitHub Advanced Security, Snyk, Semgrep, and Veracode all offer enterprise code scanning with varying levels of AI augmentation. What differentiates Claude Security's pitch is the reasoning layer — the claim that Opus 4.7 can model attacker intent rather than just match patterns — and the integration with the broader Glasswing ecosystem.

The timing is not accidental. Security researchers and threat intelligence firms have documented a surge in AI-assisted exploit development over the past 12 months. Adversaries are using large language models to accelerate the time from vulnerability discovery to weaponized exploit. Anthropic's argument is that defenders need equivalent capability, and that a model trained with safety at its core is better positioned to provide it responsibly than a general-purpose model used without guardrails.

Claude Security includes scheduled and targeted scans, integration with audit systems, and improved tracking of triaged findings. Access for Claude Team and Max customers is described as coming soon, which suggests Anthropic intends to extend the product beyond the enterprise tier — potentially putting AI-powered vulnerability scanning in front of a much broader population of developers.

What to Watch

The public beta designation means Anthropic is still collecting signal on false positive rates, scan coverage, and enterprise workflow integration. The multi-stage validation pipeline is a direct response to the credibility problem that has plagued automated security tools: if analysts cannot trust the output, they will not act on it.

Two pressure points are worth monitoring. First, the architecture-level reasoning capability will be tested against real-world adversarial codebases — modern enterprise environments are rarely clean, and the interaction effects between legacy code, third-party dependencies, and cloud infrastructure create a combinatorially complex attack surface. Second, the question of what happens when Claude Security itself becomes an attack surface — researchers have already demonstrated prompt injection vulnerabilities in AI code review agents — will remain live as adoption scales.

For now, Anthropic has moved from safety-focused AI lab to active participant in enterprise security infrastructure. The combination of Claude Security in public beta, Project Glasswing with $100 million committed, and the KPMG global alliance represents the most ambitious enterprise security play from an AI-native company to date.