Powerful AI to find hack risk, tech giants join hands

In a major step towards stronger cybersecurity, Anthropic has introduced a new AI model called Claude Mythos Preview. This advanced system is designed to detect and even exploit software weaknesses, helping experts understand how hackers might attack systems.​

Alongside, the company has launched Project Glasswing— a joint effort with top tech companies like Apple, Google, Microsoft, Amazon Web Services, and NVIDIA. The main aim is to use this powerful AI to identify and fix security flaws in important systems like banking, healthcare, power supply, and logistics.

According to a statement by Anthropic, Claude Mythos has already shown strong performance. It has reportedly found thousands of serious vulnerabilities, including very old bugs that were missed for years. Some of these include issues in operating systems, web browsers, and even widely used software tools.

This shows how powerful and effective the model is compared to previous AI systems. 

Explaining the purpose, Jared Kaplan, Chief Science Officer at Anthropic, said that the goal is to create awareness and help responsible organisations get a head start in securing both open-source and private systems. Adding to this, Logan Graham, who leads testing of new AI models, called Claude Mythos a starting point for a major industry shift, saying it could be a “reckoning” moment for how companies handle cybersecurity.

Due to safety concerns, Anthropic has not made this AI available to the public. The company believes that if such technology falls into the wrong hands, it could be misused for cyberattacks. For now, access is limited only to trusted partners and organisations working on critical infrastructure.

Project Glasswing also includes funding support, with millions of dollars being invested to improve open-source security. The goal is to stay one step ahead of cybercriminals in a time when AI is making hacking easier and faster.​