American tech giant Anthropic has introduced Project Glasswing on Tuesday, an AI tool that can analyse code and detect weaknesses in the widely used software. The tech firm says that its initiative focuses on improving the security of the software partners such as Google, Microsoft, Apple, Amazon Web Services, Cisco, CrowdStrike, and NVIDIA. How Project Glasswing WorksAnthropic wrote in its blog post that the initiative follows internal observations of its Claude Mythos2 Preview model, which easily detects weaknesses, analyses code, and suggests ways those flaws could be exploited. According to the firm, this model has identified hundreds of previously unknown vulnerabilities across operating systems, browsers, and other widely used software. The tech company
stated that its Project Glasswing will help the tech giants, allowing them to deploy the model to scan and secure both their proprietary and open-source software.
Early Tests Show AI Can Find Exploits With Little Human Help The tech company stated that its
Project Glasswing will help the tech giants, allowing them to deploy the model to scan and secure both their proprietary and open-source software. Anthropic plans to share insights from this initiative publicly, with an initial report likely within 90 days highlighting vulnerabilities addressed and improvements suggested. As per early tests conducted by the AI model, it identified and chained together exploits with minimal human input, showcasing its autonomous abilities. The blog post notes that all the disclosed vulnerabilities since then have been patched, with others to be revealed once fixes are in place.
Anthropic stated that no single firm can solve the problem of cyberattacks alone, and hence, it is a call for close cooperation between security experts, governments, and companies. The tech giant plans to share further updates in the coming months, including the vulnerabilities fixed by its AI and the lessons learned through it. The Project Glasswing has been positioned as the first step, with further collaboration expected as AI continues to evolve.