Anthropic Opens Claude Mythos Preview to Big Tech in Urgent Cybersecurity Defense Push

Anthropic has launched a new initiative aimed at strengthening cyber defenses by giving a select group of major technology and security companies early access to Claude Mythos, its unreleased and most advanced AI model.

The program, known as Project Glasswing, will allow organisations including Amazon Web Services, Apple, Microsoft, Google, Cisco, Broadcom, CrowdStrike, JPMorganChase and Nvidia to use a preview version of the model for defensive cybersecurity work. Anthropic is also extending access to around 40 additional organisations responsible for maintaining critical software infrastructure, enabling them to scan and secure both internal systems and open source code.

Anthropic says the move is driven by growing concern that Mythos could introduce unprecedented cybersecurity risks if similar capabilities fall into the wrong hands. The company believes the model’s advanced reasoning and agentic coding abilities could dramatically reshape the cybersecurity landscape, but has also warned that those same strengths may make large scale AI driven cyberattacks significantly more likely in the near future.

The existence of Claude Mythos first emerged in March, when reports revealed Anthropic was developing a model internally described as its most powerful to date. In a draft blog post that was briefly made public, the company reportedly said Mythos was already far ahead of any other AI model in cyber capabilities and could foreshadow a wave of systems able to exploit software vulnerabilities at a pace defenders may struggle to match.

That warning has already had an impact on the market. Investor concern over increasingly capable AI models and their potential effect on traditional cybersecurity products saw shares in several major security firms fall sharply following reports about Mythos and Anthropic’s broader push into security tooling.

Anthropic says the Mythos Preview has already identified thousands of zero day vulnerabilities in recent weeks, including critical flaws across major operating systems and web browsers. Some of the issues it uncovered had reportedly gone undetected for years, including a 27 year old vulnerability in OpenBSD, an operating system long regarded for its security credentials.

Despite the defensive focus of Project Glasswing, Anthropic has openly acknowledged the dual use nature of the model. A system capable of finding vulnerabilities at scale could just as easily be weaponised by malicious actors. The company has warned that frontier AI cyber capabilities are advancing far faster than the time it takes to harden global digital infrastructure, making early defensive deployment essential.

Anthropic does not currently plan to release Mythos Preview to the general public. Instead, it says the long term goal is to safely deploy Mythos class models at scale only once stronger safeguards are in place. For now, Project Glasswing is being positioned as an urgent effort to ensure powerful new AI cyber tools are used to strengthen defenses before attackers gain comparable capabilities.

Photo Credit: DepositPhotos.com