DoorDash and Snowflake adopt Claude Security

Security teams already had too many bugs to review. Now they have a new problem — AI is making it easier to find exploitable ones fast. That is the backdrop for Anthropic’s latest push: on April 30, it put Claude Security into public beta for Claude Enterprise customers, after a limited preview earlier this year. DoorDash is one of the named users, and Anthropic says the tool has already been tested by hundreds of organizations. (claude.com) ### What is Claude Security, exactly? Basically, it is Anthropic’s security-review product for codebases. It was first introduced on February 20, 2026 as “Claude Code Security” in limited research preview, then reintroduced this week as “Claude Security” in public beta. The pitch is simple: scan a codebase, find vulnerabilities, rank them, and suggest fixes that a human can review before anything ships. (anthropic.com)g this now? Because the company is openly framing cybersecurity as an AI speed race. In the beta announcement, Anthropic says current models are already highly effective at finding software flaws, and that the next generation will be better at exploiting them. It also points to its Project Glasswing work and Mythos Preview as evidence that frontier models are moving into territory that can rival elite human vulnerability researchers. (claude.com) ### What does the tool do differently? The key claim is that Claude Security is not just matching patterns the way classic static-analysis tools do. Anthropic says it tries to reason across the codebase more like a human reviewer — tracing how components interact, following data flow through the application, and looking for context-dependent flaws like business-logic bugs or broken access control. Then it runs a second pass on its own (claude.com)lse positives. (anthropic.com) ### Why does that matter to a company like DoorDash? At big companies, the hard part is rarely “did a scanner find something.” The hard part is whether the finding is real, severe, and actionable enough that an engineer will stop work and fix it. Anthropic’s beta post includes a quote from DoorDash security chief Suha Can saying the tool helps surface deep vulnerabilities accurately and route findings into workflows where engineers (anthropic.com)red. (claude.com) ### What changed in the beta release? A few practical things. Public beta brings scheduled and targeted scans, easier integration with audit systems, and better tracking for triaged findings. Anthropic also says organizations do not need to build a custom agent or wire up an API just to start using it — if they already use Claude Enterprise, the product is available directly in Claude. That lowers the activation energy a lot. (claude. ([claude.com)re do Snowflake, CrowdStrike, and Microsoft fit? This is where the original framing needs a little cleanup. DoorDash is named as a direct user in Anthropic’s own launch materials. CrowdStrike and Microsoft Security are named as technology partners embedding Opus 4.7 into their tools, not necessarily as companies using Claude Security internally in the same way. Snowflake is clearly a major Anthropic partner, but the strongest of(claude.com)ta-user quote in the launch post. (claude.com) ### So is this replacing existing security tools? Probably not. It looks more like an extra layer for the kinds of bugs rule-based tools miss. Traditional scanners are still good at well-known patterns and dependency issues. Claude Security is being positioned for the messier class of vulnerabilities that span files, services, or assumptions — the ones that usually burn analyst time. (anthropic.com)Anthropic launched another enterprise feature. It is that major software teams are being offered an AI reviewer meant to sit inside the security workflow, not beside it. If that actually reduces false positives and produces review-ready patches, the security queue gets shorter. If it does not, it becomes just another dashboard. (claude.com)