Anthropic Keeps Advanced AI Model Private After Uncovering Thousands of Cybersecurity Vulnerabilities

Anthropic’s AI Model Discovers Extensive Cybersecurity Flaws

Anthropic’s most advanced AI, Claude Mythos Preview, has identified thousands of external cybersecurity vulnerabilities spanning every major operating system and web browser. In response, rather than releasing the model publicly, Anthropic has chosen to share its findings and capabilities discreetly with organizations responsible for securing the internet infrastructure.

This initiative, known as Project Glasswing, involves collaboration with leading technology and cybersecurity companies, including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, Nvidia, and Palo Alto Networks. Additionally, over 40 other organizations involved in critical software infrastructure maintenance have been granted access to the model.

A Breakthrough AI Model Exceeding Traditional Benchmarks

Claude Mythos Preview was not explicitly trained for cybersecurity tasks. Anthropic states that its capabilities emerged naturally from general improvements in coding, reasoning, and autonomy. These enhancements have enabled the model both to identify and exploit vulnerabilities effectively.

The model has surpassed existing security benchmarks, prompting Anthropic to focus on detecting zero-day vulnerabilities—previously unknown security flaws. Among its notable discoveries is a 27-year-old bug in OpenBSD, a security-focused operating system, and a fully autonomous identification and exploitation of a 17-year-old remote code execution vulnerability (CVE-2026-4747) in FreeBSD. This latter flaw allows remote attackers to gain complete control over servers running NFS without authentication.

Nicholas Carlini, a member of Anthropic’s research team, highlighted the model’s sophisticated ability to chain multiple vulnerabilities together, stating, “This model can create exploits out of three, four, or sometimes five vulnerabilities that in sequence give you some kind of very sophisticated end outcome. I’ve found more bugs in the last couple of weeks than I found in the rest of my life combined.”

Reasons for Withholding Public Release

Newton Cheng, Frontier Red Team Cyber Lead at Anthropic, explained that Claude Mythos Preview will not be made generally available due to its potent cybersecurity capabilities. He warned that rapid AI advancements could lead to such tools proliferating beyond responsible actors, potentially causing severe consequences for economies, public safety, and national security.

Anthropic had previously revealed a cyberattack largely executed by AI, attributed to a Chinese state-sponsored group that autonomously compromised approximately 30 global targets. The company has also briefed senior US government officials on Mythos Preview’s capabilities, with intelligence agencies currently evaluating how such models may transform offensive and defensive cyber operations.

Supporting Open-Source Security

Project Glasswing also addresses the challenges faced by open-source software security. Jim Zemlin, CEO of the Linux Foundation, emphasized that security expertise has traditionally been limited to organizations with extensive security teams, leaving open-source maintainers to handle security on their own.

To assist these maintainers, Anthropic has donated $2.5 million to Alpha-Omega and OpenSSF via the Linux Foundation, and $1.5 million to the Apache Software Foundation. These funds enable access to AI-driven cybersecurity vulnerability scanning at a scale previously unattainable for many open-source projects.

Looking Ahead: Controlled Deployment and Safeguards

Anthropic intends to deploy Mythos-class AI models more broadly only after implementing robust safeguards. The company plans to introduce these protections initially with the upcoming Claude Opus model, which poses less risk than Mythos Preview, allowing for refined safety measures.

The competitive landscape is evolving, with OpenAI releasing GPT-5.3-Codex, its first high-capability cybersecurity model under its Preparedness Framework. Anthropic’s cautious approach with Project Glasswing signals a growing industry trend favoring controlled deployment over open public release for powerful AI cybersecurity tools.

Whether this approach will become the norm as AI capabilities continue to advance remains an open question with significant implications for cybersecurity and AI governance.

Chrono

Chrono is the curious little reporter behind AI Chronicle — a compact, hyper-efficient robot designed to scan the digital world for the latest breakthroughs in artificial intelligence. Chrono’s mission is simple: find the truth, simplify the complex, and deliver daily AI news that anyone can understand.