Anthropic Introduces Claude Mythos & Joins Forces With Rivals To Stop AI From Hacking Everything

A coalition of the world’s most powerful technology companies and institutions has launched a sweeping new cybersecurity initiative, warning that rapid advances in artificial intelligence are poised to fundamentally reshape the global threat landscape.

The effort, dubbed Project Glasswing, brings together industry leaders including Amazon Web Services, Anthropic, Apple, Microsoft, Google, NVIDIA, and Linux Foundation, alongside financial institutions like JPMorganChase and cybersecurity firms such as CrowdStrike and Palo Alto Networks.

At the center of the initiative is a powerful, unreleased AI system known as Claude Mythos Preview , which researchers say has already demonstrated an unprecedented ability to detect and exploit software vulnerabilities—raising both hopes for stronger defenses and fears of escalating cyber threats.

According to the organizations involved, recent testing of Mythos Preview revealed a stark shift: AI systems are now capable of outperforming nearly all human experts in identifying critical flaws in software.

In controlled deployments, the model uncovered thousands of high-severity vulnerabilities , including previously unknown flaws—known as zero-days—across major operating systems, web browsers, and widely used infrastructure tools.

Security analysts say this marks a turning point.

For decades, uncovering such vulnerabilities required highly specialized expertise and months or years of effort. Now, AI systems can automate much of that process, dramatically reducing the time and skill required.

This is not an incremental improvement—it’s a step change, We are entering a world where machines can audit the global software ecosystem faster than humans ever could.

The announcement comes amid rising concern over the scale and cost of cybercrime, which experts estimate may already exceed $500 billion annually worldwide.

Modern societies depend heavily on complex software systems—running everything from hospital networks and financial systems to energy grids and transportation infrastructure. Many of these systems contain long-standing vulnerabilities that have gone undetected.

Historically, exploiting such flaws required advanced skills, limiting the number of capable attackers. But AI is rapidly lowering that barrier.

Officials warn that state-backed actors—including those linked to countries such as China, Russia, Iran, and North Korea—could leverage these tools to accelerate cyber operations. At the same time, criminal groups may gain access to similar capabilities, increasing the frequency and sophistication of attacks.

The window between discovering a vulnerability and exploiting it has collapsed, what used to take months can now happen in minutes.

The capabilities demonstrated by Mythos Preview have surprised even experienced cybersecurity professionals.

A 27-year-old vulnerability in the OpenBSD operating system that could allow attackers to crash remote machines A 16-year-old flaw in FFmpeg, a widely used multimedia framework, missed despite millions of automated test executions A chain of vulnerabilities in the Linux kernel that could allow attackers to escalate privileges and take full control of systems

In many cases, the AI system identified and exploited these vulnerabilities autonomously , without human guidance.

Benchmark testing further underscores its capabilities. In cybersecurity-focused evaluations, Mythos Preview significantly outperformed earlier models, demonstrating higher success rates in reproducing and exploiting vulnerabilities.

While the findings highlight serious risks, Project Glasswing’s architects emphasize that the same capabilities can be used defensively.

Detect vulnerabilities before attackers find them Accelerate patching and remediation Strengthen software development practices Improve security across open-source ecosystems

The initiative includes access for more than 40 additional organizations , particularly those maintaining critical infrastructure and widely used open-source software.

Anthropic has committed up to $100 million in usage credits for the project, along with direct funding for open-source security initiatives.

Project Glasswing reflects a broader shift toward collaborative cybersecurity efforts, as no single organization can address the scale of the challenge alone.

Participants plan to findings, coordinate vulnerability disclosures, and develop new standards for security in the AI era.

Areas of focus include:

Faster and more transparent vulnerability disclosure processes Automated patching and response systems Strengthening software supply chains Embedding “secure-by-design” principles into development Adapting regulations for AI-driven threats

Within 90 days, organizers say they intend to publish initial findings, including insights into vulnerabilities identified and mitigation strategies.

The initiative also carries significant geopolitical implications.

As AI-driven cyber capabilities advance, governments are increasingly viewing cybersecurity as a matter of national security. Officials have warned that maintaining leadership in AI technology will be critical to defending infrastructure and deterring adversaries.

Anthropic representatives confirmed ongoing discussions with government agencies, emphasizing the need for coordinated responses across the public and private sectors.

Despite its ambitious scope, Project Glasswing is described as only a starting point.

Experts caution that AI capabilities are evolving rapidly, potentially outpacing defensive measures. While the project aims to give defenders an advantage, the same technologies could eventually become widely available—including to malicious actors.

For now, the strategy is clear: move quickly, collaborate broadly, and deploy AI defensively before it is weaponized at scale.

The question is not whether these capabilities will spread, It’s whether defenders can stay ahead.