// news · agents · alignment2026-05-27source: anthropic / project glasswing / security industry

Project Glasswing expands with Claude Security public beta plus cyber-verification tools for eligible security teams — agent meets SOC

Anthropic's Project Glasswing expanded this cycle with the Claude Security public beta and cyber-verification tools for eligible security teams. The Glasswing umbrella was previewed earlier in 2026 as the lab's defensive-cyber positioning; the public-beta milestone moves the program from research artifact to operational tool. Security operations centers running Claude in defensive workflows are now a deployment category.

The Claude Security public-beta scope is the operational news. The toolset includes agent-driven vulnerability triage, log-analysis assistance against SIEM and EDR feeds, threat-intel correlation across the customer's existing detection surface, and incident-response playbook execution under analyst supervision. Each individual capability has equivalents in the existing security-tooling market; the integration into a single Claude-mediated surface is what makes the offering different. The cyber-verification tools — software supply-chain attestation, automated security-control validation, configuration-drift detection — extend the agent's reach into preventive controls rather than just detection-and-response.

The eligibility-gating is the policy piece. Cyber-defense capability is dual-use; an agent fluent enough to assist defenders is by construction capable of assisting attackers. Anthropic's eligibility process for the cyber-verification tools is the lab's procedural answer to the dual-use problem: vetted customers get access, vetting requires demonstrating a legitimate defensive use case and accepting use-policy constraints. The model is similar to how the largest cloud providers handle restricted-capability features in regulated-industry sales. Combined with the Claude Managed Agents sandbox tier, the broader Glasswing umbrella positions Claude as the agent platform regulated security teams can deploy without violating their internal governance.

See our analysis →

Anthropic — Project Glasswing expansion and Claude Security beta → · Anthropic Alignment — Defensive cyber and Mythos preview research notes → · Dark Reading — AI in defensive cyber 2026 trends →