Anthropic has expanded its Project Glasswing initiative to 150 organizations around the globe, the company said Wednesday. The program uses artificial intelligence to help protect critical software and infrastructure from cyber threats.
What Project Glasswing Does
Project Glasswing focuses on applying AI tools to harden the security of essential systems — think power grids, financial networks, hospital databases. Instead of relying purely on rule-based defenses, the AI continuously analyzes traffic patterns, flags anomalies, and suggests fixes in near real-time. The idea is to keep attackers out before they can lock down a system or steal data.
Anthropic has built the project around its proprietary language models, though the company hasn't detailed exactly how those models are fine-tuned for security work. What is clear is that Glasswing isn't a product you buy off the shelf. Organizations join the program through a direct partnership, getting access to Anthropic's research team and ongoing model updates.
The Expansion
The jump to 150 participating organizations marks a significant scale-up. Just a year ago, the project counted fewer than three dozen members, most of them in North America. Now the roster spans financial firms, utilities, cloud providers, and government contractors across Europe, Asia, and the Middle East.
Anthropic didn't release a full list of participants, but it did say that roughly a third of the new additions come from the energy sector. Another quarter are in healthcare, where ransomware attacks have become a persistent problem. The remaining organizations are split between finance, logistics, and telecommunications.
Why Now?
Cyberattacks on critical infrastructure have climbed sharply over the past two years. Colonial Pipeline, JBS Foods, and the Irish health service are just a few high-profile targets that suffered major disruptions. Existing security tools often fail because they can't adapt fast enough to novel attack patterns — something AI can potentially do, provided it's trained on the right data.
Anthropic says Project Glasswing has already helped block several attempted intrusions that traditional antivirus or firewall software missed. The company shared no specifics, citing confidentiality agreements. But it noted that the system catches roughly 40% more suspicious behaviors than conventional methods, according to internal testing.
The expansion isn't just about numbers. Anthropic also plans to open-source some of Glasswing's core detection models later this year, which would let smaller organizations — those without the resources for a full partnership — benefit from the same AI protections.
No timeline has been set for that release, and Anthropic hasn't said whether the open-source models will be free or licensed. What's certain is that the demand for proactive, AI-driven security isn't slowing down.
