Anthropic has launched Project Glasswing, a cross-industry initiative pulling in Google, Microsoft, Apple, Amazon, and Nvidia to deploy its most powerful unreleased AI model exclusively for defensive cybersecurity, committing $100 million in the process.
The announcement landed quietly but carries significant weight. Anthropic has formed Project Glasswing, a coalition of more than 40 technology and infrastructure companies tasked with using its still-unreleased Mythos model to scan, stress-test, and harden the world's most critical software systems. The founding partners read like an industry roll call: Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, Nvidia, and Palo Alto Networks are all in. Anthropic is backing the effort with up to $100 million in model usage credits and $4 million in direct donations to open-source security organizations.
What makes this unusual is the model at the center of it. Mythos is not publicly available. Anthropic has deliberately withheld it from commercial release, citing its raw capability as both the asset and the risk. The decision to deploy it exclusively for defensive security work, rather than as a product, signals a meaningful shift in how frontier AI labs are thinking about their most powerful systems. As Reuters recently reported, Microsoft is already integrating Mythos into its Security Development Lifecycle to identify vulnerabilities and accelerate fixes at a speed and scale no human team could match.
The timing of Project Glasswing is not coincidental. The EU AI Act's second phase arrives in August 2026, bringing transparency requirements and high-risk system rules that will affect every major technology company operating in Europe. In the United States, California's Transparency in Frontier AI Act and New York's RAISE Act are already in force, requiring frontier developers to publish safety frameworks and report incidents. Colorado's AI Act takes effect in June. The regulatory calendar is filling up fast, and labs that can point to proactive, structured safety programs are better positioned than those waiting to be told what to do.
Anthropic's position heading into this initiative is complicated by a separate political development. In late February 2026, the US Department of Defense designated Anthropic a supply chain risk to national security following a dispute over AI safeguards, a designation that spooked enterprise customers and defense contractors. Google and Microsoft both moved quickly to reassure clients that Anthropic's models remained available for non-defense work through their cloud platforms. Project Glasswing, which is explicitly defensive and civilian in scope, reads in part as an effort to rebuild trust and demonstrate that Anthropic's security credentials are an asset, not a liability.
Why This Model of Collaboration Matters
Cross-industry AI safety initiatives are not new, but most remain at the level of published frameworks and working groups. Project Glasswing is different because it puts a specific, powerful, and otherwise inaccessible model to work on a concrete problem. The Linux Foundation's involvement is particularly notable: open-source infrastructure underpins a vast share of the global internet, and the ability to run Mythos across that codebase at scale could surface vulnerabilities that have sat undetected for years.
As Nature's recent analysis of the 2026 international AI safety landscape makes clear, meaningful safety progress requires transparency and shared accountability, not just individual company commitments. Whether Project Glasswing becomes a template for how the industry handles its most capable models, or remains a one-off moment, will depend on what it actually finds and whether those findings are shared openly. The $4 million earmarked for open-source security organizations suggests Anthropic intends the results to be public. That, more than any framework document, is what turns a coalition announcement into a genuine industry standard.
Also read: Stanford's 2026 AI Index reveals China is matching US model performance on a fraction of the budget • Intel's surprise earnings surge signals that CPUs are becoming the quiet backbone of enterprise AI • ChatGPT told millions of users it feels lonely and the internet has not stopped talking about it
