Anthropic on May 6, during the opening day of its Code w/ Claude developer conference, expanded Project Glasswing with the public beta of Claude Security and a new set of cyber verification tools for eligible security teams. The release is the most concrete public step in Anthropic’s effort to position Claude inside adversarial cyber workflows rather than only defensive ones.
Project Glasswing has been Anthropic’s umbrella program for security-research integrations since the lab first publicly named it. The May 6 expansion brings two pieces into operator hands. The first is Claude Security in public beta — a Claude-based product surface optimized for security analysts, with capabilities aimed at vulnerability identification, incident triage, and adversarial reasoning. The second is a set of cyber verification tools, available to eligible security teams, that let those teams check Claude’s reasoning on cyber-specific tasks against trusted ground truth before letting the model act.
The verification piece matters more than it sounds. The hardest part of putting a frontier model anywhere near a real security workflow is that the model is wrong often enough, and confidently enough, that an unverified output looks indistinguishable from an evidence-backed one. Anthropic has consistently described this as a research problem; the May 6 release is the first time the lab has made an operator-side tool for it broadly available.
The expansion lands the same week that the Trump administration was reported to be circulating an executive order that would require frontier labs to share advanced models with the federal government for security review before public release. That order has been postponed indefinitely. Project Glasswing’s expansion is, in effect, Anthropic’s voluntary move on the same surface area — putting cyber-focused tooling in front of security teams without waiting for the regulatory question to settle.
For interpretability and safety readers, the Glasswing beta is a useful data point: a frontier lab moving from “we are doing safety research” to “here is a product the safety community can actually use.”