Anthropic has announced Project Glasswing, a new industry initiative to secure critical software using its unreleased frontier model, Claude Mythos Preview. The project addresses the dual-use risks of advanced AI in cybersecurity.
It’s a collaborative effort involving major tech and security organizations to use Claude Mythos Preview defensively. The goal is to proactively find and patch vulnerabilities in foundational software; operating systems, browsers, open-source projects, etc. before malicious actors can exploit them.
Launch partners include: Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, Linux Foundation, Microsoft, NVIDIA, Palo Alto Networks and Anthropic itself. Anthropic is also extending access to roughly 40 additional organizations responsible for critical software infrastructure.
The company is committing up to $100 million in usage credits for the model and $4 million in direct donations to open-source security efforts including via the Linux Foundation. Claude Mythos Preview is Anthropic’s most capable model yet for coding and agentic tasks. It demonstrates a major leap in software security capabilities.
Register for Tekedia Mini-MBA edition 20 (June 8 – Sept 5, 2026).
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Register for Tekedia AI Lab.
It can find and exploit vulnerabilities better than all but the most skilled human experts. In testing, it reportedly identified thousands of high-severity zero-day vulnerabilities across every major OS, web browser, and other critical software. One notable example: It surfaced a long-standing bug in OpenBSD; a security-focused OS that had persisted for ~27 years.
Anthropic describes this as a stark fact: Frontier AI models have reached the point where they can surpass top humans at vulnerability research. Because of the offensive potential, the model is not being released publicly—it’s initially limited to trusted partners in Project Glasswing for defensive use only.
Pricing for participants is set at $25/$125 per million input and output tokens via major cloud platforms. This follows a recent accidental leak of information about Mythos, which highlighted its step change in capabilities and raised cybersecurity concerns. By giving defenders early access, the initiative aims to close vulnerability gaps faster than attackers can weaponize similar AI capabilities.
Anthropic plans to share learnings broadly so the entire ecosystem benefits. It highlights the evolving AI-cybersecurity arms race. Stronger AI coding agents could automate bug hunting at unprecedented scale, but the same tools could enable more sophisticated attacks if they proliferate unchecked. Project Glasswing is framed as an urgent response to prepare for this new era.
Anthropic is choosing controlled access over open release for this model, echoing past decisions like OpenAI’s early GPT-2 handling. Reactions note that this could reshape cybersecurity practices, with some experts calling it unsettling because it signals AI is now powerful enough in offensive cyber tasks to warrant restricted access. Cybersecurity stocks reportedly rose following the news.
Project Glasswing is positioned as a starting point, not a complete solution—Anthropic emphasizes that industry, open source, researchers, and governments all need to collaborate. This is a notable development in how frontier labs are handling powerful, dual-use capabilities—prioritizing defense for critical infrastructure while limiting broader exposure.
The AI arms race refers to the intense global competition—primarily between the United States and China, but also among leading tech companies—to develop, deploy, and dominate advanced artificial intelligence technologies. This rivalry spans frontier models, infrastructure like chips, data centers, energy, military applications, economic productivity, and dual-use capabilities like cybersecurity.
The US leads in cutting-edge frontier model capabilities like models from OpenAI, Anthropic, and Google, with a reported edge of several months over Chinese counterparts. China, however, excels in rapid industrial deployment, manufacturing integration, and scaling AI across sectors like logistics, robotics, and consumer applications.
Beijing invests heavily in AI as national infrastructure, while the US emphasizes scaling compute and pursuing AGI-like breakthroughs. Export controls on advanced chips remain a flashpoint, though recent policy shifts have sparked debate. Tech giants like NVIDIA, Microsoft, Google, Meta, Amazon, Anthropic, etc. pour hundreds of billions into AI infrastructure, models, and applications.