Anthropic’s Project Glasswing: What It Means for AppSec

An AI just found critical vulnerabilities that survived decades of human review. If your security program isn’t built for this moment, it’s already behind.

Surprise! An AI just did what your security team couldn’t.

Last week, Anthropic announced that Claude Mythos Preview — their shiny new frontier model — autonomously found a 27-year-old vulnerability in OpenBSD, a 16-year-old flaw in FFmpeg hiding in a line of code that automated tools had run five million times without catching it, and casually chained together Linux kernel vulnerabilities to achieve full machine takeover. No human help. No hand-holding. Just an AI doing what the best security researchers in the world do — only faster, cheaper, and without needing coffee.

Oh, and it did this across every major operating system and every major web browser.

This is Project Glasswing — Anthropic’s move to put these capabilities to work for defense, alongside AWS, Microsoft, Google, Apple, Cisco, CrowdStrike, NVIDIA, and JPMorganChase, before the bad guys get there first.

Bold strategy. Also, probably overdue.

The old playbook is broken. Project Glasswing just proved it.

Here’s a fun fact nobody in the industry wants to say out loud: the reason so many critical vulnerabilities have survived for decades isn’t because they were cleverly hidden. It’s because finding them was hard, expensive, and required expertise that most teams simply don’t have.

That friction? Gone. AI just dissolved it.

CrowdStrike noted in the Glasswing announcement that the window between vulnerability discovery and exploitation has collapsed from months to minutes. That’s not hyperbole — it’s the operational reality that Mythos Preview just demonstrated at scale.

The threat environment has moved on. Most security programs haven’t. That’s not an accusation — it’s just math.

If your program is still built around slow-cycle scanning, manual triage, and a remediation queue measured in weeks, the question isn’t whether you’re exposed. It’s how long before someone notices.

Open source is the most dangerous blind spot you have right now

Let’s talk about the part nobody wants to admit.

The vulnerabilities Mythos Preview found weren’t buried in some obscure proprietary codebase. They were in OpenBSD. FFmpeg. The Linux kernel. Software that runs firewalls, servers, video pipelines, and critical infrastructure that the entire modern world depends on — including, almost certainly, your applications.

Open source has always been the largest and least defended part of the software supply chain. Most maintainers don’t have dedicated security teams. They’re running on community goodwill and caffeine. They rely on manual review and the hope that someone will catch what they missed.

Spoiler alert: AI just caught what everyone missed. For 27 years!

Project Glasswing’s $4M in donations to open source security organizations is a good start. But let’s be honest — it’s a band-aid on a very large wound. The vast majority of open source software, including the transitive dependencies buried three layers deep in your applications, remains largely unprotected.

If you don’t have continuous, deep visibility into your full open source dependency tree, you have blind spots that an AI-powered attacker can now exploit faster than your team can respond. And they will.

This is now a national security conversation, and Washington is listening

It’s not a coincidence that Project Glasswing landed when it did.

Anthropic has been increasingly vocal in Washington about AI’s role in national security, and the Glasswing announcement makes the government dimension explicit. It calls out state-sponsored attacks from China, Iran, North Korea, and Russia by name, frames AI-powered cyber capabilities as a top priority for democratic nations, and confirms that Anthropic is in active discussions with US government officials about Mythos Preview’s offensive and defensive implications.

Translation: the adults in the room have noticed, and regulations are coming.

For CISOs in regulated industries, federal contracting, financial services, or critical infrastructure — the organizations that wait for a mandate to modernize their security programs will be well behind the ones that moved proactively. Governments don’t telegraph this stuff for fun. When they start naming adversaries and holding meetings, it’s time to pay attention.

The same AI that threatens you is your best defense, but only if you use it

Here’s the plot twist you’ve been waiting for.

The capabilities that make Mythos Preview terrifying in the wrong hands are exactly what defenders need. AI that can autonomously find decades-old vulnerabilities across millions of lines of code can just as easily find them in your codebase, before an attacker does.

This is not a future state. Microsoft, AWS, Google, CrowdStrike, and Palo Alto Networks are already running Mythos Preview against their own systems. They didn’t join Project Glasswing for the press release. They joined because they understand that the only way to fight AI-powered offense is with AI-powered defense.

The question isn’t whether to use AI for security. It’s whether you’re already doing it — or whether you’re going to wait until you really, really have to.

What Mend.io does differently

The whole point of Glasswing is that, AI-powered offense is here. The only credible response is AI-powered defense — and that’s exactly what we’ve been building.

Agentic SCA and SAST: When we say agentic, we mean it. Mend.io connects directly to agentic development tools via our MCP Server, feeding real-time vulnerability data from both SCA and SAST so the tool itself can autonomously create remediations — no human in the loop, no ticket filed and forgotten. The vulnerability is found, the fix is generated, and it happens in the flow of development, not weeks later in a backlog.

AI-based remediation: For teams not yet operating in fully agentic environments, Mend.io’s AI-based remediation automatically generates, validates, and prioritizes fixes — so your team is closing the gap at the speed this threat environment demands, not drowning in a queue while the exposure window stays wide open.

And as AI models become first-class components of your software supply chain, Mend AI ensures they don’t become your next blind spot — tracking every model, framework, and agent in your applications with the same rigor you already apply to open source.

The takeaway

Anthropic didn’t assemble the world’s most influential technology companies for a fun collaborative exercise. They did it because a threshold has been crossed, the threat is real, and the old ways of doing security are no longer sufficient.

The organizations that come out ahead won’t be the ones that read about this moment. They’ll be the ones that used it as the forcing function to finally build AppSec programs that are fast enough, smart enough, and automated enough to match what’s now possible on the attack side.

So. Are you ready or are you still working through that backlog?

Mend.io helps security and development teams find and fix vulnerabilities at the speed and scale modern software demands. Talk to our team about how Mend.io can help you respond to this new threat landscape.