Anthropic Launches Project Glasswing to Secure Critical Softwares

Anthropic Launches Project Glasswing to Secure Critical Software as AI Finds Thousands of Vulnerabilities

Project Glasswing brings together major technology firms including Amazon, Apple, Google, Microsoft, Nvidia and Cisco to strengthen cybersecurity using advanced AI capabilities.

SAN FRANCISCO: Anthropic has announced Project Glasswing, a new cybersecurity initiative aimed at protecting critical software and digital infrastructure as artificial intelligence becomes increasingly capable of identifying and exploiting software vulnerabilities.

The project brings together some of the world’s largest technology and security companies, including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, Nvidia, and Palo Alto Networks.

According to Anthropic, the initiative was created in response to the capabilities demonstrated by Claude Mythos Preview, an unreleased frontier AI model that the company says can identify software vulnerabilities at a level that surpasses most human security researchers.

AI Security Capabilities Reach New Level

Anthropic said Claude Mythos Preview has already discovered thousands of previously unknown high-severity vulnerabilities, including flaws affecting major operating systems, web browsers, and other critical software infrastructure.

The company warned that rapid advances in AI could dramatically lower the expertise required to find and exploit security weaknesses, potentially creating new risks for governments, businesses, and critical infrastructure operators.

“AI models have reached a level of coding capability where they can surpass all but the most skilled humans at finding and exploiting software vulnerabilities,” Anthropic said in its announcement.

Vulnerabilities Found in Major Software

Among the examples highlighted by Anthropic:

• A 27-year-old vulnerability discovered in OpenBSD, a security-focused operating system widely used in critical infrastructure.
• A 16-year-old flaw found in FFmpeg, one of the world’s most widely used video processing frameworks.
• Multiple vulnerabilities identified and chained together within the Linux kernel, potentially allowing attackers to gain full control of affected systems.

Anthropic said the vulnerabilities were responsibly disclosed and have since been patched by the relevant software maintainers.

Industry Giants Join the Effort

As part of Project Glasswing, participating organizations will use Claude Mythos Preview to strengthen their own security operations and identify weaknesses in critical systems.

Several companies involved in the initiative described AI-driven cybersecurity as a major shift for the industry.

Cisco Chief Security and Trust Officer Anthony Grieco said AI capabilities have crossed a threshold that fundamentally changes how organizations must approach infrastructure protection.

Microsoft Executive Vice President of Cybersecurity and Microsoft Research Igor Tsyganskiy said the initiative provides an opportunity to identify and mitigate risks earlier while improving security outcomes at scale.

CrowdStrike Chief Technology Officer Elia Zaitsev warned that the window between vulnerability discovery and exploitation is shrinking rapidly as AI capabilities improve.

$100 Million Commitment for Cyber Defense

Anthropic is committing up to $100 million in usage credits for Claude Mythos Preview to support Project Glasswing participants and other organizations responsible for critical software infrastructure.

The company has also pledged $4 million in direct donations to open-source security organizations, including the Linux Foundation’s Alpha-Omega and OpenSSF initiatives, as well as the Apache Software Foundation.

More than 40 additional organizations responsible for maintaining critical software and open-source infrastructure have also been granted access to the model through the program.

Performance Claims

Anthropic says Claude Mythos Preview significantly outperforms its previous-generation models across cybersecurity, coding, reasoning, and software engineering benchmarks.

The company noted that the model’s advanced coding abilities are a key factor behind its cybersecurity performance, allowing it to identify vulnerabilities that have remained hidden despite years of human review and automated testing.

However, Anthropic said it currently has no plans to release Claude Mythos Preview publicly, citing concerns about the risks associated with such powerful cyber capabilities.

Looking Ahead

Project Glasswing is expected to continue for many months, with Anthropic planning to publish updates on lessons learned, vulnerabilities fixed, and recommendations for improving cybersecurity practices in an AI-driven world.

The company says future work will focus on areas including vulnerability disclosure, software update processes, open-source security, secure software development practices, automated patching, and cybersecurity standards for regulated industries.

As AI systems become more powerful, Anthropic argues that collaboration between technology companies, governments, researchers, and open-source communities will be essential to ensure defensive cybersecurity capabilities stay ahead of potential threats.

Key Takeaway: Project Glasswing represents one of the largest coordinated efforts yet to use advanced AI for defensive cybersecurity, as industry leaders prepare for a future where artificial intelligence can discover software vulnerabilities at unprecedented speed and scale.
Read more from Tech Categories

Related posts:

US Expands AI Safety Testing Before Public Release of New Models Pope Leo XIV Urges AI ‘Disarmament’ in Major Vatican Warning on Technology Risks Claude Opus 4.8 Launches With Dynamic Workflows and Faster AI Performance Nvidia ramps up Taiwan spending to $150bn a year as AI race intensifies Microsoft Scout