Anthropic’s Claude Mythos, Project Glasswing, and How to Handle the Coming Patch Tsunami

One of this week’s most alarming and significant announcements came from Anthropic, who revealed that their newest AI model, Claude Mythos, delivers unexpected and substantial improvements in its ability to discover and create working exploits for software vulnerabilities. While AI models have been applied to this problem for years, Mythos represents enough of a step-function improvement that Anthropic chose not only to keep it private temporarily, but to also launch Project Glasswing, an industry initiative involving leading technology and security vendors, as well as open source communities. The goal of Project Glasswing is to proactively apply Mythos’ capabilities to major software components, giving the industry a window of time to discover and patch vulnerabilities before those same capabilities can be weaponized by malicious actors.

I applaud this. It reflects a level of responsibility that we should expect, but don’t always get, in our industry. At the same time, I want to be clear-eyed about its implications, because I think the security industry risks treating this as reassuring news when it should be treated as a red alert. If you’re responsible for information security in an enterprise, this is the moment to act. First, I’ll explain the why, then talk about the how.

What Project Glasswing Will Actually Look Like in Practice

During the Claude Mythos preview window, participating vendors and their ecosystems will use Claude Mythos to discover hundreds of vulnerabilities and release updates to address them. That’s genuinely good. But let’s trace what happens next.

First, the entire software supply chain will simultaneously have to update component libraries, test, and release new versions. And then, the onus of applying those patches will fall entirely on enterprises. All those new versions of all that softsare, across both commercial and open source software, then need to work their way through enterprise change management and update processes.

So, ask yourself this question right now: if 80% or more of your OS, IT, and application software simultaneously release new patches, how long will that take to fully apply across your environment? 

And what about your organization’s custom software? If the majority of those systems depend on libraries that need to be updated, what does your build, test, and release process look like at that scale?

It’s not hyperbole to describe what’s coming as a patch tsunami. It will be a significant, sustained effort across the enterprise, disruptive to other work in progress, and for some legacy or custom software it may prove extremely difficult or functionally impossible to apply security updates at all.

One more factor deserves honest acknowledgment: even with the breathing room that Project Glasswing provides, it will not identify and remediate 100% of vulnerabilities. Adversaries’ AI models will continue to improve, and they will discover and exploit latent vulnerabilities. This window of respite is real, but it is finite, and we need to be prepared for what comes after. 

Ain’t I a ray of sunshine?

Now that I’ve made your day considerably less pleasant, let me offer a constructive path forward.

Patching vulnerable software is not your only possible response here, and I’d argue it shouldn’t be your primary one. What matters more, and more urgently, is adopting a default-deny posture and applying Zero Trust security controls across your environment.

Here’s the logic: if we operate from the assumption that essentially all our systems are vulnerable, and will continue to be vulnerable regardless of our patching efforts, then the most effective defensive move is to ensure that those systems are invisible and inaccessible to unauthorized actors in the first place. Given a choice between patching a vulnerable system and cloaking it from unauthorized access, I’d take the cloak. Ideally, you do both, but if you have to prioritize, default-deny, enforced network access control is the more durable and scalable approach.

The good news is that there are well-established architectural and technology approaches to achieving this, including free and open source options. It requires time and deliberate effort, but the security outcomes are meaningfully better than patch management alone. Zero Trust, done well, assumes breach and removes implicit trust, which is precisely the posture this moment demands.

The window that Project Glasswing provides is valuable. Don’t spend it only chasing patches. Spend it building the kind of architecture that makes your exposure to any given vulnerability significantly smaller.Want to talk through how to apply this to your enterprise? Contact us. We’d be glad to help you think this through.