Anthropic moved its Claude Security product into public beta this week, making it available to all Claude Enterprise customers. The tool uses Claude Opus 4.7 to scan codebases for software vulnerabilities and generate patches — and early results from the preview phase suggest it’s finding things that traditional security tooling has been missing for years.

What Claude Security Does

The core capability is codebase scanning. Security teams can point Claude Security at a full GitHub repository or a specific directory, and it will work through the code the way a human researcher would: tracing data flows, reading source files, mapping the interactions between components, and synthesising a picture of how vulnerabilities could be exploited.

Before flagging anything to an analyst, the system assigns a confidence rating to each finding. The intent is to reduce the noise problem that makes so many security tools exhausting to use in practice — analysts spend more time triaging false positives than fixing real problems.

The product started life in February 2026 as “Claude Code Security” in a limited research preview. According to Anthropic, hundreds of organisations used it during that phase and collectively found exploits in production code that other tools had missed entirely, in some cases for years.

The Vendor Ecosystem Around It

Anthropic is not positioning Claude Security as a standalone replacement for existing security platforms. Technology partners including CrowdStrike, Palo Alto Networks, SentinelOne, Wiz, and Trend Micro’s TrendAI are integrating Claude Opus 4.7 directly into their own platforms. That means the AI reasoning capabilities can run inside tools security teams are already using, rather than requiring a separate workflow.

On the services side, Accenture, BCG, Deloitte, Infosys, and PwC are deploying Claude Security for enterprise clients as part of broader security posture engagements.

Claude Team and Max customers will get access soon, Anthropic said, though no specific date was given.

Why AI Is the Right Tool for This Problem

The timing reflects where the threat landscape is heading. AI is now being used by attackers to discover and exploit vulnerabilities faster than most organisations can respond. Traditional static analysis tools scan for known patterns; they struggle with complex multi-file interactions where a vulnerability only materialises when you trace logic across a full system.

Claude’s approach — reading code holistically, mapping dependencies, reasoning about how components interact — is better suited to the class of vulnerabilities that pattern-matching tools miss. The confidence rating system adds a layer of quality control that makes the output usable without overwhelming analysts.

What This Means for Business

Security teams in larger organisations will want to evaluate this during the beta period for two reasons. First, the preview data suggests real discovery value in areas existing tools underperform. Second, the partner integrations with CrowdStrike, Palo Alto, and others mean adoption does not require ripping out existing tooling.

For business leaders thinking about their AI strategy more broadly, this is worth watching as a template. Enterprise AI is maturing past “chat interface for employees” into dedicated tooling for specific, high-stakes professional workflows. Security is one of the clearest fits because the consequences of a missed vulnerability are measurable and severe.

If your team is building internal tools, running custom applications, or managing any production software — the gap between “AI that helps you write code” and “AI that actively audits what you’ve built” is closing fast. That’s a meaningful shift for risk management.

---

The practical next step is the free Working With Claude field guide. Thirty-two pages covering the ecosystem, Claude Code, and how to govern a rollout properly. Get your copy.