Anthropic Unveils ‘Project Glasswing’ to Test New AI Model With Apple, Google, and Amazon

Anthropic has launched a new initiative aimed at strengthening cybersecurity through advanced artificial intelligence, partnering with major technology players to test an unreleased model designed to identify software vulnerabilities.

The program, called Project Glasswing, gives selected partners—including Apple, Google, and Amazon—early access to a preview version of its new model, Claude Mythos Preview. The initiative is intended to evaluate the system’s capabilities in real-world environments before a broader rollout.

According to Anthropic, the model is designed to detect vulnerabilities in operating systems and web browsers with a level of accuracy that rivals, and in some cases exceeds, that of highly skilled human experts. The focus is on proactively identifying security weaknesses that could be exploited in increasingly complex digital infrastructures.

Through the partnership, participating companies will provide structured feedback on the model’s performance, helping refine its capabilities and reliability. Anthropic indicated that insights gathered during this phase may be shared with the broader industry, signaling a collaborative approach to cybersecurity innovation.

The move comes as demand grows for AI-driven security solutions, particularly as cyber threats become more sophisticated and widespread. Large technology firms are increasingly investing in advanced defensive tools to protect critical systems, user data, and global digital infrastructure.

While the model remains in a testing phase, Project Glasswing highlights a broader trend of collaboration between AI developers and major tech companies to accelerate the deployment of next-generation security technologies.