Anthropic opens Claude Security beta

Security teams already have too many bugs to chase. The new problem is speed — AI is getting good enough to find subtle flaws in code, and soon good enough to exploit them much faster than human attackers can. That makes the old security workflow feel slow and brittle. Anthropic’s move this week is basically an attempt to wedge Claude directly into that gap. ### What changed this week? Anthropic put Claude Security into public beta on April 30, 2026, and opened it to all Claude Enterprise customers. This is the product Anthropic first introduced in February as “Claude Code Security” in a limited research preview. Now it has a broader name, broader rollout, and a clearer pitch: use Claude Opus 4.7 to scan codebases for vulnerabilities and generate proposed patches without making teams build custom agents or wire up an API first. ### What does Claude Security actually do? At the core, it reads code more like a human security researcher than a rules engine. Traditional static analysis tools are great at spotting known bad patterns — hardcoded secrets, stale crypto, obvious mistakes. But they often miss logic bugs and context-dependent flaws. Anthropic says Claude Security traces how components interact, follows data through an application, validates its own findings and ships automatically. Developers still decide what gets patched. ### What’s new in the beta version? The beta adds the unglamorous enterprise stuff that usually determines whether a product gets used or ignored. Anthropic says customers now get scheduled scans, targeted scans, easier audit-system integration, and better tracking for triaged findings. That matters because security teams do not want one more isolated dashboard. They want something that fits into the ticketing, audit, and review processes they already run. ### Why are the partner names a big deal? Because Anthropic is not trying to sell this as a standalone security console only. It is also pushing Opus 4.7 into tools enterprises already trust. Anthropic named CrowdStrike, Microsoft Security, Palo Alto Networks, SentinelOne, TrendAI, and Wiz as technology partners embedding the model into their own products. It also named Accenture, BCG, Deloitte, Infosys, and PwC as services partners helping as much as a product launch. ### Why now? Anthropic’s own framing is pretty blunt: AI is compressing the timeline between finding a vulnerability and exploiting it. Earlier in April, the company launched Project Glasswing with partners including CrowdStrike, Microsoft, and Palo Alto Networks, plus access to Claude Mythos Preview for defensive security work. Claude Security is the more broadly available version of the same basic thesis — defenders need frontier models in their hands before attackers industrialize them. ### Is this just code scanning with nicer marketing? Not really — but the catch is that it still lives inside human review. That is deliberate. Anthropic has been careful to frame these tools as defender-facing systems, not autonomous exploit machines. The company’s February preview stressed that the same capabilities that help defenders could also help attackers, which is why the product includes verification steps, confidence ratings, and approval gates. The proposition is “faster expert triage,” not “let the model loose in production.” ### What matters most here? The important shift is where Anthropic wants Claude to live. Not just in a chat window, and not just as raw API access, but inside the boring infrastructure of enterprise security work. If that sticks, Claude Security is less a feature launch than a land grab — a bid to become part of the default security stack before AI-driven offense gets much stronger.