Project Glasswing: Anthropic Unites Apple, Google, Microsoft on AI Cybersecurity

In a move that would have seemed unlikely just a few years ago, some of the world’s biggest technology rivals, including Apple, Google, and Microsoft, are joining forces to address what they see as rapidly escalating cybersecurity risks driven by artificial intelligence.

The partnership, called Project Glasswing and led by Anthropic, brings together more than 45 organizations across the tech, finance, and cybersecurity sectors. Their shared goal is to use powerful next-generation AI models to identify and fix critical software vulnerabilities before they can be exploited.

The project centers around Claude Mythos Preview, an unreleased frontier AI model developed by Anthropic that has already demonstrated an ability to uncover thousands of previously unknown security flaws, including some embedded in critical infrastructure software for decades.

A new kind of security tool

Early testing suggests the results are impressive.

Mythos Preview has reportedly already discovered subtle vulnerabilities in widely used software systems, including a 27-year-old bug in the security-focused OpenBSD operating system and a long-overlooked flaw in video software that had passed automated security testing millions of times.

What makes Mythos Preview unique is the model’s ability to analyze how software behaves within larger technology ecosystems. Its ability to understand complex software systems enables it to detect weaknesses that emerge only when multiple systems interact, which is an area where traditional testing often falls short.

It seems that when it comes to cybersecurity, AI is becoming powerful enough not only to assist human security researchers but also to potentially outperform them in certain areas.

Preparing for an AI cybersecurity arms race

One of the biggest concerns driving Project Glasswing is speed. Security experts increasingly worry that AI is dramatically shortening the time between a vulnerability’s discovery and its weaponization.

This creates a dangerous dynamic, as the same AI capabilities that help defenders identify risks can also be used by attackers to scale up cyber operations. As a result, companies are under pressure to modernize their defenses as quickly as AI advances.

Project Glasswing includes more than 45 organizations spanning cloud computing, semiconductors, cybersecurity, finance, and open-source software. Participants include Amazon Web Services, Cisco, Nvidia, the Linux Foundation, JPMorgan Chase, CrowdStrike, and Palo Alto Networks. Together, they will gain controlled access to Mythos Preview to test their own systems and strengthen defenses.

This effort reflects a growing belief that no single company can handle this shift alone.

The double-edged sword of AI security

AI models capable of advanced coding and reasoning are increasingly able to both find vulnerabilities and propose ways to exploit them, meaning the tools that help defenders today could also enable attackers if they fall into the wrong hands.

Of course, concerns exist about releasing such a powerful system too broadly, as this could give attackers the same advantage that defenders are trying to build. Reports indicate that Mythos Preview has already identified high-severity vulnerabilities across major operating systems and web browsers, in some cases autonomously developing potential exploit chains. 

This capability is one reason Anthropic is restricting access to trusted partners rather than releasing the model publicly. Part of the plan is to give these organizations controlled access to Mythos Preview so they can test their own infrastructure and fix weaknesses before similar tools become widely available.

The fact that direct competitors are willing to collaborate at all goes to show how seriously the industry is taking the threat.

Supporting the open-source ecosystem

Another main focus of the project is open-source software, which supports much of today’s digital infrastructure but is often maintained by small teams or individual developers with limited resources.

To help address this gap and support the broader software ecosystem, Anthropic has also pledged $4 million in direct funding and is providing AI usage credits to support security improvements in critical open-source projects. The objective is to make advanced security tools available not only to large corporations but also to independent developers, enabling them to identify and fix security issues more quickly.

Government interest is also increasing. Anthropic confirmed it has briefed US officials on Mythos Preview’s defensive and potential offensive cyber capabilities, underscoring the national security implications of advanced AI cybersecurity tools.

A sign of what’s to come

Project Glasswing may represent a turning point in how the tech industry approaches shared security risks. It is too soon to tell right now whether it will become a lasting model for cooperation or just an early experiment in AI-driven cyber defense. 

Still, it shows that in the age of AI-accelerated threats, even the fiercest competitors may need to work together on a shared battlefield to protect the digital foundations they all depend on.

Also read: Google is expanding Gemini across search, devices, and developer tools, underscoring how quickly major tech companies are embedding advanced AI across their ecosystems.