Claude Security beta launches, giving enterprises new controls to lock down data and model access

Enterprise AI security is getting more concrete. Anthropic didn’t just add another admin toggle or compliance badge — it turned a research-y cybersecurity capability into a product enterprises can actually buy and use. On April 30, Anthropic put Claude Security into public beta for Claude Enterprise customers, with the pitch that security teams can scan big codebases, surface vulnerabilities, and generate proposed fixes without wiring up custom agents or API workflows. ### What is Claude Security? It’s an AI-assisted code security tool. Anthropic first introduced it on February 20 as “Claude Code Security” in a limited research preview for Enterprise and Team customers, plus some open-source maintainers. The new move is broader availability and a cleaner product identity — now just Claude Security, in public beta for Claude Enterprise customers. Basically, it scans software code for vulnerabilities and proposes patches for humans to review. Anthropic’s angle is that the model is not just matching known bad patterns the way classic static-analysis tools often do. It says Claude reasons through how components interact, how data moves through an application, and where business-logic or access-control flaws might hide. That matters for automated scanners to miss. ### Why is this different from a normal scanner? The pitch is “reasoning,” not signatures. Traditional tools are good at catching exposed secrets, stale crypto, and other known-pattern issues. But they can struggle when the bug depends on context spread across multiple files or services. Anthropic says Claude re-checks its own findings in multiple stages, tries to disprove them, and assigns severity and confidence — the thing security teams complain about most with automated scanners. ### What changed in the beta launch? The public beta adds the enterprise plumbing. Anthropic says Claude Security now includes scheduled and targeted scans, easier integration with audit systems, and improved tracking for triaged findings. The other big change is distribution: no custom build is required if a company already uses Claude, and Anthropic is also pushing the same capability through partners like CrowdStrike and Wiz, plus service firms including Accenture, BCG, Deloitte, Infosys, and PwC. ### Which model is underneath it? Claude Opus 4.7. Anthropic launched that model on April 16 and positions it as its strongest generally available model for coding, agents, and complex enterprise work. Opus 4.7 also carries a 1 million token context window, which helps explain why Anthropic thinks it can reason across large codebases instead of just small snippets. In plain English — the model can hold more of the software system in view at once. ### Why launch this now? Because Anthropic thinks AI is compressing the gap between finding a flaw and exploiting it. That theme runs through both the February preview and the April beta post. The company has also been talking publicly about stronger internal cyber capabilities, including Mythos Preview, which it says can match or beat elite human experts at finding and exploiting software vulnerabilities. Claude Security is the ability in defenders’ hands before attackers get there first. ### What’s the catch? The catch is that a tool that helps defenders can also hint at what strong offensive AI might look like. Anthropic is trying to thread that needle by keeping humans in the loop — nothing gets applied automatically — and by routing the product through enterprise plans and established security partners. But the broader message is pretty stark: frontier model companies are no longer just selling chatbots. They’re starting to sell security labor. ### Bottom line? Claude Security matters less as a feature launch than as a category signal. Anthropic is packaging cutting-edge model capability into a direct enterprise security product, and doing it at the moment when AI-powered bug hunting is starting to look like a real platform shift, not a demo.