Project Glasswing: the largest AI cybersecurity coalition ever formed wants to protect the most critical software in the world
Cybersecurity has never been a static field, but what is happening right now is unlike anything we have seen before.
AI models have reached a level of technical capability where they can identify and exploit vulnerabilities in software with a precision that surpasses virtually any human expert, and that changes the game irreversibly.
This is exactly the context in which Project Glasswing emerges, an initiative that brings together some of the biggest names in global technology, including Amazon Web Services, Apple, Cisco, Google, Microsoft, NVIDIA, CrowdStrike, JPMorganChase, Palo Alto Networks, Broadcom, the Linux Foundation, and Anthropic, all united by a common goal: to use that power to defend, not attack.
At the heart of this initiative is Claude Mythos Preview, a frontier model not yet publicly released by Anthropic that has already found thousands of critical zero-day vulnerabilities in virtually every major operating system and browser in the world.
Some of these flaws had survived for more than two decades without being detected, passing through millions of automated tests without leaving a trace.
What this reveals is simple and a little scary at the same time: traditional security methods have very clear limits, and artificial intelligence has already surpassed those limits a long time ago. 😬
In this article, you will learn how Glasswing works in practice, which historical vulnerabilities the Mythos Preview has already found, what the partners are saying about the initiative, the investment involved, and most importantly, why the timing of all this matters just as much as the technology itself.
How Project Glasswing Works in Practice
Project Glasswing is not simply another code scanning tool. The proposal goes far beyond that. The initiative combines the advanced reasoning power of Anthropic’s AI models with the deep technical knowledge of the biggest players in the global tech sector, creating a collaborative ecosystem where artificial intelligence acts as a top-tier security researcher capable of analyzing massive codebases with a speed and depth that would be humanly impossible to replicate.
The model does not just scan code looking for known failure patterns. It reasons about system behavior, anticipates attack vectors, and identifies combinations of conditions that only become dangerous under very specific scenarios, exactly the kind of thing that slips past traditional scanners.
Claude Mythos Preview was developed with a specific focus on offensive and defensive security tasks, which means it was trained to think like an experienced security researcher, not just a syntax checker. It can cross-reference information across different parts of a system, understand complex dependencies, and identify where a seemingly minor flaw can become a critical entry point when combined with other environmental conditions. This is something that conventional static and dynamic analysis methods simply cannot do with the same efficiency.
Another important point is that the project does not operate in isolation. Each partner brings a unique set of infrastructure, data, and technical expertise that continuously feeds and refines the analysis process. AWS contributes processing capacity at scale, Google and Microsoft bring deep knowledge of their own software ecosystems, while companies like CrowdStrike and Palo Alto Networks add real-world threat context and observed malicious actor behavior from the field. This combination of cutting-edge artificial intelligence with specialized human knowledge is what makes Glasswing truly different from anything that existed before in the cybersecurity space.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
Beyond the main partners, Anthropic has also extended access to Mythos Preview to a group of more than 40 additional organizations that build or maintain critical software infrastructure. These organizations can use the model to scan and protect both proprietary and open-source systems, significantly expanding the defensive reach of the initiative.
Vulnerabilities That Survived for Decades
One of the most impressive and, honestly, most disturbing aspects of what Project Glasswing has delivered so far is the discovery of vulnerabilities that were hidden in critical systems for decades. We are talking about flaws that went through security audits, millions of automated test runs, code reviews by extremely qualified engineers, and still remained invisible throughout all that time.
Anthropic shared three concrete examples that illustrate the gravity of the situation well:
- OpenBSD, 27-year-old flaw: Mythos Preview found a vulnerability in OpenBSD, which has a reputation as one of the most secure operating systems in the world and is widely used to run firewalls and other critical infrastructure. The flaw allowed an attacker to remotely crash any machine running the system simply by connecting to it.
- FFmpeg, 16-year-old flaw: The model discovered a vulnerability in FFmpeg, the library used by countless software applications for encoding and decoding video. The flaw was in a line of code that automated testing tools had executed five million times without ever detecting the problem.
- Linux Kernel, autonomous exploit chain: Mythos Preview found and chained together multiple vulnerabilities in the Linux kernel, the software that runs most of the servers in the world, in a fully autonomous manner, allowing an attacker to escalate from regular user access to full control of the machine.
All of these vulnerabilities have already been reported to the maintainers of the respective software and have received fixes. For many other flaws still in the patching process, Anthropic published a cryptographic hash of the details on the Frontier Red Team blog, committing to disclose the details as soon as fixes become available. 🔍
What makes these discoveries even more relevant is the type of vulnerability found. These are not trivial errors or configuration issues. They are structural flaws that, under certain conditions, allow remote code execution, privilege escalation, or sensitive data leakage. Some of them require a very specific chain of conditions to be exploited, which explains why they went unnoticed for so long. A traditional scanner cannot simulate the reasoning needed to realize that the combination of conditions A and B in a certain context creates an opening that, in isolation, would never represent a risk. The AI model does exactly that: it connects dots that were separated and sees the full picture.
From a practical standpoint, this capability represents a pretty significant shift in how the industry will need to think about the secure software lifecycle. If an AI model can find in weeks what humans did not find in decades, the inevitable question is: what else is still out there, hidden, waiting to be discovered by someone with bad intentions?
The Bigger Picture: Why This Matters Now
To understand the urgency of Project Glasswing, you need to look at the broader landscape of global cybersecurity. The software we all use every day, responsible for running banking systems, storing medical records, connecting logistics networks, keeping power grids running, and much more, has always contained bugs. Many are minor, but some are serious security flaws that, if discovered by malicious actors, can enable system hijacking, operational disruption, or sensitive data theft.
The global financial costs of cybercrime are hard to estimate precisely, but they hover around 500 billion dollars per year, according to recent estimates. Attacks on corporate networks, healthcare systems, energy infrastructure, transportation hubs, and government agencies have repeatedly demonstrated the severe consequences these flaws can cause.
On the geopolitical front, state-sponsored attacks from countries like China, Iran, North Korea, and Russia have threatened to compromise the infrastructure that supports both civilian life and the military readiness of various nations. Even smaller-scale attacks, like those targeting individual hospitals or schools, can still cause substantial economic damage, expose sensitive data, and even put lives at risk.
Over the past year, AI models have become increasingly effective at reading and reasoning about code, demonstrating a remarkable ability to identify vulnerabilities and find ways to exploit them. Claude Mythos Preview represents a leap in these cyber capabilities. Ten years after the first DARPA Cyber Grand Challenge, frontier AI models are now becoming competitive with the best humans at discovering and exploiting vulnerabilities.
The good news is that the same capabilities that make AI models dangerous in the wrong hands make them invaluable for finding and fixing flaws in important software, and for producing new software with far fewer security bugs from the start.
What the Partners Are Saying
The commitment of such diverse companies to Project Glasswing is no accident, and the public statements from partners make it clear there is a genuine sense of urgency behind the initiative.
Cisco emphasized that AI capabilities have crossed a threshold that fundamentally changes the urgency needed to protect critical infrastructure against cyber threats, and that there is no going back. According to the company, working with these models showed that it is possible to identify and fix security vulnerabilities in hardware and software at a pace and scale that was previously impossible.
AWS mentioned that its teams analyze more than 400 trillion network flows per day looking for threats, and that AI is central to the ability to defend at scale. The company has been testing Claude Mythos Preview in its own security operations, applying it to critical codebases where the model is already helping to harden code.
Microsoft, represented by Igor Tsyganskiy, Executive Vice President of Cybersecurity and Microsoft Research, pointed out that we are entering a phase where cybersecurity is no longer limited by purely human capability. When tested against CTI-REALM, Microsoft’s open-source code security benchmark, Claude Mythos Preview showed substantial improvements compared to previous models.
CrowdStrike was quite direct: the window between a vulnerability being discovered and being exploited by an adversary has collapsed. What used to take months now happens in minutes with AI. That is not a reason to slow down, it is a reason to move together and faster.
The Linux Foundation brought a fundamental perspective on open source, highlighting that open-source software maintainers, whose work underpins much of the world’s critical infrastructure, have historically been left on their own when it comes to security. Glasswing offers a concrete path to change that equation, making AI-augmented security an accessible resource for all maintainers, not just those who can afford expensive security teams.
JPMorganChase, represented by Pat Opet, Chief Information Security Officer, reinforced that advancing cybersecurity and the resilience of the financial system is central to the bank’s mission, and that Glasswing offers a unique opportunity to evaluate next-generation AI tools for defensive cybersecurity on critical infrastructure.
Google mentioned its own AI-powered tools, such as Big Sleep and CodeMender, built to find and fix critical software flaws, reinforcing that the industry needs to work together on emerging security issues.
Palo Alto Networks was especially emphatic in stating that Mythos Preview identified complex vulnerabilities that previous-generation models simply missed entirely, and that everyone needs to prepare for AI-assisted attackers because there will be more attacks, faster attacks, and more sophisticated attacks. 🌐
The Investment Behind Glasswing
Project Glasswing is not just a statement of intent. Anthropic is committing up to 100 million dollars in usage credits for Claude Mythos Preview for the partners and additional participants in the initiative. This amount covers substantial usage during the research phase. After that period, the model will be available to participants at a cost of 25 dollars per million input tokens and 125 dollars per million output tokens, accessible via Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.
Beyond the usage credits, Anthropic made direct donations to open-source security organizations:
- 2.5 million dollars to Alpha-Omega and OpenSSF, through the Linux Foundation
- 1.5 million dollars to the Apache Software Foundation
These resources will allow open-source software maintainers to respond to the evolving threat landscape. Interested maintainers can sign up through the Claude for Open Source program.
Plans for the Future and Next Steps
Today’s announcement is just the beginning of a long-term effort. Project Glasswing partners will receive access to Claude Mythos Preview to find and fix vulnerabilities in their foundational systems, which represent a very large portion of the shared global cyber attack surface. The work is expected to focus on tasks such as local vulnerability detection, black-box binary testing, endpoint protection, and penetration testing of systems.
Anthropic has also outlined a clear roadmap for transparency and collaboration:
- Within 90 days, Anthropic will publish a report on what was learned, the vulnerabilities fixed, and the improvements that can be disclosed
- Partners will share information and best practices with each other to the extent possible
- In collaboration with leading security organizations, a set of practical recommendations will be produced on how security practices should evolve in the AI era
These recommendations may include vulnerability disclosure processes, software update processes, open-source and supply chain security, secure development lifecycle practices, standards for regulated industries, scalability and triage automation, and patch automation.
One point worth highlighting: Anthropic does not plan to make Claude Mythos Preview available to the general public. The eventual goal is to allow models of this class to be deployed at scale safely, but that depends on advances in safeguards that detect and block the most dangerous outputs of the model. The company plans to launch new safeguards with an upcoming Claude Opus model, allowing it to improve and refine these protections with a model that does not present the same level of risk as Mythos Preview.
The Geopolitical Dimension and the Role of Governments
Anthropic also revealed that it is in ongoing discussions with United States government authorities about Claude Mythos Preview and its offensive and defensive cyber capabilities. Protecting critical infrastructure is a national security priority for democratic countries, and the emergence of these cyber capabilities is yet another reason why the U.S. and its allies need to maintain a decisive advantage in AI technology.
Governments play an essential role both in maintaining that advantage and in evaluating and mitigating the national security risks associated with AI models. Anthropic stated it is ready to work with local, state, and federal representatives to assist in these tasks.
The long-term vision is that Glasswing can seed an even larger effort involving industry and the public sector, with all parties helping to address the biggest questions about the impact of powerful models on security. In the medium term, an independent, third-party body capable of bringing together organizations from both the private and public sectors may be the ideal space for continuing this work on large-scale cybersecurity projects.
What This Means for the Future of Cybersecurity
Project Glasswing represents a pretty clear inflection point in the field of cybersecurity. The ability of AI models to find and exploit vulnerabilities is advancing at a pace that makes preventive action not just desirable, but absolutely necessary. The window of opportunity for defenders to maintain an edge over potential attackers is narrowing rapidly, and initiatives like this one exist precisely to ensure that the defensive side gets there first.
The fact that companies that normally compete fiercely with each other decided to collaborate on this project says a lot about the gravity of the situation. When AWS, Google, Microsoft, and Apple are sitting at the same table to solve a problem, it is because the problem is big enough to make competition secondary.
At the end of the day, Glasswing is not just about finding bugs in software. It is about redefining how the world thinks about digital security in an era where artificial intelligence is simultaneously the greatest threat and the best defense available. And if the early results are any indication of what is to come, the next few months promise to be pretty interesting for anyone following this space. 🚀
