Apple and Anthropic launch cybersecurity project Apfelpatient

Anthropic has launched a cybersecurity initiative called "Project Glasswing," with Apple as a partner. The new AI model, Claude Mythos, has already discovered thousands of serious vulnerabilities in all major operating systems and browsers.

AI is increasingly being used not only for programming but also for securing software. Anthropic – the company behind the AI platform Claude – has unveiled a new initiative called Project Glasswing, which aims precisely at this. Apple, along with Amazon, Google, Microsoft, Nvidia, and other technology companies, is among the founding partners. At its core is a new AI model, not yet publicly available: Claude Mythos.

Project Glasswing has a clear goal: to make the world's most important software more secure before high-performance AI models fall into the hands of potentially malicious actors. Anthropic is providing a preview, developed by Claude Mythos, to selected partners – including Apple, AWS, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, Nvidia, and Palo Alto Networks. Over 40 other organizations that develop or maintain critical software have also been granted access.

What Claude Mythos has already found

The results are remarkable: According to Anthropic, Claude Mythos has already identified thousands of serious security vulnerabilities – in every major operating system and every major web browser. Some of these vulnerabilities had survived decades of human review and millions of automated security tests undetected.

As a concrete example, Anthropic cites a vulnerability in the Linux kernel that Claude Mythos not only found but also linked with other vulnerabilities – with the result that an attacker could theoretically have gained complete control over a machine.

Why Apple is involved

Apple's participation in Project Glasswing is relevant for several reasons. The company operates multiple operating systems—macOS, iOS, iPadOS, watchOS, tvOS, and visionOS—which together run on over two billion devices. Security vulnerabilities in these systems can have far-reaching consequences, ranging from data loss and surveillance to financial damage.

Apple has previously relied on AI-powered security testing and operates its own bug bounty program, rewarding external security researchers for reporting vulnerabilities. The collaboration with Anthropic expands this arsenal with an AI model that appears capable of finding vulnerabilities missed by previous methods.

Claude Mythos: More than Cybersecurity

Claude Mythos is not exclusively designed for security research. According to Anthropic, the model also shows significant improvements over the previous top model, Claude Opus 4.6, in the areas of reasoning, agent-based search, computer use, and especially agent-based coding. Anthropic has published a system card with detailed benchmarks.

However, Anthropic currently has no plans to make Claude Mythos publicly available. The long-term goal is to enable the secure and scalable deployment of models of this performance class – not only for cybersecurity, but for a wide range of applications.

What this means for Apple users

There will be no immediate direct impact on Apple products. However, in the long term, the initiative could help ensure that security vulnerabilities in iOS, macOS, and their associated frameworks are discovered and patched more quickly – before attackers can exploit them. In an era where AI tools are also used offensively, the defensive use of such models is a logical and necessary step. (Image: Shutterstock / Stockinq)