Anthropic ships Claude Security beta

Security teams already have scanners. The problem is that most scanners are great at known patterns and much worse at weird, high-impact bugs that live in business logic or tangled permissions. That gap matters more now, because AI can help defenders find flaws faster — but it can also help attackers do the same. Anthropic’s news on April 30 is that Claude Security is now in public beta for Claude Enterprise customers, with the company turning its earlier Claude Code Security preview into a broader defensive product. (business-standard.com) ### What is Claude Security, exactly? Basically, it is Anthropic’s security-focused code review system. It scans a company’s codebase for vulnerabilities, tries to verify whether a suspected issue is real, assigns severity, and then proposes patches for people to inspect. Anthropic is pitching it as something security teams can start using inside Claude Enterprise without building custom agents or wiring up separate API workflows first. (infosecurity-magazine.com) ### Why isn’t this just another static analyzer? Static analysis tools usually work like rulebooks — they match code against known bad patterns. That catches obvious mistakes, but it can miss bugs that depend on how parts of an application interact. Anthropic’s pitch is that Claude reads code more like a human security researcher would, tracing data flow an(infosecurity-magazine.com) here. Not “AI scans code,” but “AI can reason about the code path.” (anthropic.com) ### What changed this week? The big change is productization. Anthropic first released this capability in February 2026 as a limited research preview under the name Claude Code Security. Now it has been renamed Claude Security and opened as a public beta for Claude Enterprise customers. That shift matters because it moves the tool from “interesting controlled experiment” to “something Anthropic wants companies to operationalize.” (anthropic.com) ### Why does the human review part matter? Because false positives are poison in security. If a tool floods analysts with junk, the team stops trusting it. Anthropic says Claude re-checks its own findings in a multi-stage verification process, adds confidence and severity ratings, and leaves the final decision to the customer. The patch does not get applied automatically. Your team still decid(anthropic.com)footnote — it is the safety valve. (anthropic.com) ### Why pair this with TrendAI? Turns out Anthropic is not just shipping a feature. It is also helping build an ecosystem around security use cases for Claude Opus 4.7. TrendAI, the enterprise AI security arm of Trend Micro, said on April 30 that it is collaborating with Anthropic to expand use of Claude Opus 4.7 for security research and risk reduction. In plain English, Anthropic finds bugs (anthropic.com)y work to actual enterprise remediation and threat workflows. (prnewswire.com) ### Why is Anthropic leaning into this now? Because the company has been publicly arguing both sides of the cyber-AI story. Anthropic has shown off stronger cyber capabilities in newer models and has also warned that the same jump in capability can help offensive actors. So Claude Se(prnewswire.com)at model risk alone. (anthropic.com) ### What’s the real takeaway? Enterprise AI stacks are starting to grow a native security layer. For a while, copilots mostly meant coding help, search, and chat. Now vendors want those same models to sit inside vulnerability discovery, triage, and patching loops. The catch is that security buyers will care less about the demo and more about precision, workflow fit, and whether the tool finds (anthropic.com)infosecurity-magazine.com)