← BACK_TO_LOG

Anthropic's Project Glasswing gets its first update

2026-05-24 · anthropic

Anthropic's May 22 update shows Project Glasswing moving from launch into execution. The company says its partners and testers are already using Claude Mythos Preview to find and triage large numbers of high-severity issues across critical software. For developers and security teams, the important signal is that frontier models are being positioned as practical defense tools, not just research demos. Anthropic is also linking the effort to productized workflows, including Claude Security and a cyber verification program.

Key Features or Updates

Anthropic says the first month of Project Glasswing has produced more than ten thousand vulnerabilities across partner systems, with a strong share landing in the high- or critical-severity range. The update also describes how Mythos Preview is being used on open-source projects and why the bottleneck has shifted from discovery to verification, disclosure, and patching. Anthropic is pairing the research with tools and programs meant to help security teams operationalize the findings.

Impact on Developers

The practical impact is that AI-assisted security work is starting to look like a real production workflow rather than a lab exercise. If the model can surface issues faster than humans can process them, engineering teams need sharper triage habits, tighter disclosure discipline, and faster patch cycles. That changes how teams think about both internal code quality and third-party dependency risk.

How to use it

Anthropic points Enterprise customers to Claude Security for codebase scanning and proposed fixes, and it is also opening a cyber verification program for legitimate security work. Teams should treat the announcement as a reason to review their own patch cadence, disclosure process, and vulnerability intake pipeline. The right first step is to test the tooling on a contained codebase and make sure the human review bottleneck is explicit.

Read Original Post →