US-based technology company Anthropic on Tuesday unveiled Project Glasswing, an AI-powered tool designed to review code and identify vulnerabilities in widely used software. The company says the initiative is aimed at strengthening the security of software used by its partners, including Google, Microsoft, Apple, Amazon Web Services, Cisco, CrowdStrike, and NVIDIA.
In a blog post, Anthropic said the project is based on insights from its Claude Mythos2 Preview model, which can quickly identify weaknesses, analyse code, and suggest how those flaws could be exploited. The company claims the model has already found hundreds of previously unknown vulnerabilities in operating systems, browsers, and other widely used software.
Project Glasswing is designed to help major tech companies use this AI tool to scan and secure both their own and open-source software.
Anthropic highlighted that addressing cyber threats will require close collaboration between companies, governments, and security experts, positioning Project Glasswing as an initial step toward broader cooperation as AI continues to evolve. The company also plans to share further updates in the coming months.
The initiative is designed to help major tech firms use AI to scan and secure both proprietary and open-source software. Anthropic said it will publish its findings, with an initial report expected within 90 days outlining vulnerabilities and suggested improvements.
In early tests, the AI demonstrated the ability to identify and connect exploits with minimal human input, showcasing its autonomy. All vulnerabilities discovered so far have been patched, while others will be disclosed once fixes are in place.
