Anthropic has unveiled Project Glasswing, a new security initiative built around a frontier AI model that the company says has reached an alarming level of capability in finding and exploiting software vulnerabilities. The effort brings together some of the biggest names in cloud, software, hardware, finance and cybersecurity in what amounts to an early attempt to keep defensive security ahead of the next wave of AI-driven cyber threats.
The scale of the partner list immediately signals how seriously the issue is being treated. Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA and Palo Alto Networks are all joining Anthropic as launch partners. Their involvement suggests a growing view inside the industry that the next major leap in AI capability may have as much to do with security disruption as with productivity or automation.
At the center of the project is Claude Mythos Preview, an unreleased general-purpose model that Anthropic says is already operating at a level where it can outperform nearly all human experts at discovering and exploiting software weaknesses. That claim, if it holds up in practice, would mark a major shift in the cybersecurity landscape.
The initiative is built around a defensive race
Project Glasswing is based on a simple but urgent idea: if frontier AI models are becoming powerful enough to identify serious vulnerabilities at scale, then those capabilities need to be placed in the hands of defenders before they spread more widely to unsafe actors. Anthropic says Mythos Preview has already found thousands of high-severity vulnerabilities, including issues affecting major operating systems and web browsers.
The company’s concern is not only what the model can do today, but how quickly similar capabilities may proliferate. In that scenario, the time between vulnerability discovery and real-world exploitation could shrink dramatically, raising the risk to critical infrastructure, public safety and national security. Project Glasswing is therefore being framed less as a normal product launch and more as a defensive mobilization.
The message is clear: the industry may be entering a phase where cyber defense is no longer limited by human scale, and waiting for broader deployment before responding could be a strategic mistake.
Major companies will use the model in real security work
As part of the initiative, the launch partners will use Mythos Preview in defensive security work across their own critical systems. Anthropic says the model will be applied to tasks such as finding vulnerabilities, probing software weaknesses and helping secure foundational infrastructure that large numbers of users and organizations rely on every day.
The company also says it has extended access to more than 40 additional organizations that build or maintain critical software infrastructure, allowing them to scan both first-party and open-source systems. That move broadens the effort beyond a small circle of elite technology firms and pushes the initiative toward the wider software ecosystem, including the open-source foundations on which much of the digital economy depends.
Anthropic says it will share what it learns from these efforts so that the wider industry can benefit, suggesting the project is intended to produce not only immediate security gains, but also new operating standards and best practices for the AI era.
Anthropic is backing the effort with money and access
To support the initiative, Anthropic is committing up to 100 million dollars in usage credits for Mythos Preview and 4 million dollars in direct donations to open-source security organizations. That financial commitment is substantial enough to show that the company sees the project as a long-term strategic effort rather than a limited pilot.
The funding also matters because many of the most important software maintainers in the world do not have the resources of large technology platforms. If frontier AI is going to change the balance of cybersecurity, then open-source maintainers and smaller infrastructure players will need tools and support as well, not just the largest corporate defenders.
By pairing elite partner access with broader support for critical software ecosystems, Anthropic is trying to position Project Glasswing as both a high-end security collaboration and a wider industry preparedness effort.
This is also a warning about where AI is heading
The deeper significance of Project Glasswing is that it functions as both an announcement and a warning. Anthropic is effectively saying that frontier AI has crossed a threshold where software security can no longer be treated as a secondary issue. If models can now discover and exploit vulnerabilities at a level beyond almost every human practitioner, then the balance between attackers and defenders may soon shift in unpredictable ways.
That is why the company is emphasizing urgency. It argues that defending global cyber infrastructure may take years of coordination across AI developers, software companies, security researchers, open-source maintainers and governments, while frontier model capabilities may continue advancing in a matter of months. That mismatch in speed is what makes the current moment so consequential.
Project Glasswing is therefore not being presented as a complete solution. It is a starting point. But it also marks something more important: a public acknowledgment from one of the leading AI labs that the security implications of frontier models are no longer theoretical. They are already strong enough to force the industry into collective action.
