Anthropic has launched a brand new cybersecurity initiative designed to discover how superior AI can assist determine and handle software program vulnerabilities throughout the worldwide know-how ecosystem.
This system, often known as Venture Glasswing, brings collectively main know-how corporations like Microsoft and AWS, together with organizations answerable for sustaining essential digital infrastructure, to check a sophisticated, unreleased AI system constructed particularly for safety evaluation.
On the heart of the undertaking is a frontier AI mannequin referred to as Claude Mythos, which partnering corporations can entry and use. It reportedly has robust coding and reasoning capabilities that may analyze complicated software program environments and determine potential vulnerabilities.
“We’ve been testing Claude Mythos Preview in our personal safety operations, the place it’s already serving to us strengthen our code,”
AWS mentioned in response to its participation in this system.
The initiative displays rising concern throughout the cybersecurity trade in regards to the speedy evolution of AI capabilities. As fashions grow to be higher at analyzing code and figuring out weaknesses, specialists warn that each defenders and attackers may achieve highly effective new instruments. Anthropic says Venture Glasswing is a response to that new actuality.
Venture Glasswing is designed to discover how these capabilities may be deployed responsibly to strengthen defensive safety practices earlier than such methods grow to be broadly accessible.
Inside Venture Glasswing and the Claude Mythos Mannequin
Claude Mythos, not like Anthropic’s publicly out there fashions, has not been launched as a result of issues in regards to the dangers related to its capabilities. As a substitute, entry is restricted to a rigorously chosen group of organizations throughout the know-how and cybersecurity sectors.
Regardless of this, Anthropic says the mannequin has already demonstrated a exceptional means to determine software program vulnerabilities amongst taking part corporations. Throughout inner testing, Mythos reportedly found 1000’s of high-severity flaws, together with vulnerabilities affecting main working methods and broadly used internet browsers.
In lots of instances, these flaws had remained undetected regardless of years of safety analysis. The mannequin has additionally demonstrated a capability to generate working exploits for sure vulnerabilities.
In response to Anthropic’s analysis, Mythos efficiently developed working exploits in 181 makes an attempt throughout testing, whereas earlier fashions resembling Claude Opus 4.6 confirmed nearly no success in the identical benchmark.
“When examined towards CTI-REALM, our open-source safety benchmark, Claude Mythos Preview confirmed substantial enhancements in comparison with earlier fashions,”
Igor Tsyganskiy, International CISO and EVP of Safety and Analysis at Microsoft, mentioned.
Anthropic famous that these capabilities weren’t explicitly skilled into the mannequin however as a substitute emerged from enhancements in coding means, reasoning, and autonomy.
Participation within the initiative consists of greater than 40 organizations answerable for constructing or sustaining key software program infrastructure. Corporations concerned embrace Amazon Internet Providers, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Basis, Microsoft, Nvidia, and Palo Alto Networks. Anthropic has additionally dedicated as much as $100 million in utilization credit for the undertaking alongside $4 million in funding for open-source safety efforts.
A Strategic Bid to Outline AI’s Function in Cyber Protection
Whereas Venture Glasswing focuses on vulnerability discovery, the broader significance of the initiative lies in how Anthropic is positioning itself inside the cybersecurity ecosystem.
The undertaking indicators an ambition to maneuver past being solely an AI mannequin developer and towards turning into a foundational supplier of safety infrastructure powered by frontier AI.
The size of the initiative displays that ambition. By partnering with main know-how corporations, cloud suppliers, and infrastructure maintainers, Anthropic is constructing a community able to testing and deploying AI-driven defensive capabilities throughout giant parts of the worldwide software program ecosystem.
By working immediately with these actors, Anthropic is integrating its fashions into the operational layers of cybersecurity reasonably than treating them as standalone instruments.
Taken collectively, the size of the initiative, the associate roster, and the numerous monetary commitments recommend a broader purpose: Anthropic is just not merely experimenting with AI-powered vulnerability discovery; it’s making an attempt to place itself as an organization able to shaping how frontier AI is built-in into the defensive safety ecosystem.
AI because the Subsequent Layer of Cybersecurity Infrastructure
Venture Glasswing highlights how Anthropic is making an attempt to broaden its position past that of an AI mannequin developer. By embedding its know-how inside organizations answerable for sustaining essential software program infrastructure, the corporate is positioning itself nearer to the operational core of cybersecurity.
The initiative locations Anthropic’s fashions immediately within the fingers of corporations that construct and safe a lot of the world’s digital ecosystem. Cloud suppliers, chipmakers, enterprise software program distributors, and open-source maintainers all play a task in sustaining the methods that underpin fashionable computing.
By working with these actors, Anthropic goals to make its AI an important layer of an organization’s tech stack reasonably than a bolt-on layer like a copilot.
The monetary scale of the undertaking reinforces that ambition. With as much as $100 million in utilization credit and direct funding for open-source safety initiatives, Anthropic is investing closely in constructing long-term relationships with the communities and organizations that keep foundational software program. The corporate goals to be a sustained participant within the cybersecurity ecosystem reasonably than merely providing AI instruments from the sidelines.
