SonicWall joins Anthropic's Project Glasswing to find flaws

Here's a dynamic that doesn't get talked about enough: attackers are already using AI. They're using it to discover vulnerabilities faster, write exploits at scale and compress the window between disclosure and active exploitation down to days, sometimes hours. The offensive side of the AI arms race didn't wait for anyone to feel ready. Threat actors started using AI as fast as they could, for whatever they could, and people and businesses alike have been paying the price. 

But the defensive side has been catching up. And that gap is exactly what Project Glasswing was built to close.

SonicWall has joined Anthropic's Project Glasswing, an industry initiative that gives participating security teams access to Claude Mythos 5, a frontier AI model not available to the general public, for defensive security research. SonicWall is applying it directly to vulnerability discovery and code review across its own software, with one goal: find and fix security issues before an attacker exploits them. 

Why This Matters for the People Defending the Network

The reality here is a little uncomfortable. 

Vulnerabilities move from disclosure to exploitation faster than most patch cycles can keep up with. That's not a new problem, but AI has made it significantly worse on the attacker's side. Automated scanning, AI-assisted exploit development and shared tooling across threat actor communities mean that a newly disclosed CVE can be weaponized at scale before many organizations have even read the advisory. That's not fear-mongering, it's just the situation the security industry finds itself in.

Security vendors aren't exempt from that pressure. SonicWall's products sit at the edge of networks for over 17,000 partners: MSPs, MSSPs, and the small and mid-sized businesses (SMBs) they protect. A vulnerability in that infrastructure isn't an abstract risk. It's a direct line into the organizations that depend on it.

Participating in Project Glasswing means SonicWall's security team can bring the same category of AI capability to its defensive testing workflows that attackers are already applying offensively. That's not a nice-to-have. It's the new baseline for what responsible software security looks like.

The Part About Sharing

What makes Glasswing distinct from just getting access to a powerful model is the commitment that comes with it. Participants don't just use the program for their own benefit. They feed findings and best practices back to Anthropic to help refine the framework for the broader security community.

That matters because defenders have historically been at a structural disadvantage when it comes to sharing. Attackers share tools and techniques openly. Defenders, for competitive and legal reasons, often don't. Glasswing is a structured attempt to change that dynamic, and SonicWall's participation means its learnings from working across thousands of partner environments contribute to something bigger than any single vendor's security posture.

The goal isn't just to find vulnerabilities in SonicWall's own code. It's to help build a model for how AI-assisted defensive security actually works at scale and make sure that knowledge gets back to the people who need it.

The times are changing. We have to change with them. 

More information about Project Glasswing is available here.