Anthropic and Claude Mythos: AI Revolutionizes Cybersecurity

⚡

Key Takeaways

1Anthropic has launched Claude Mythos Preview, an AI model designed to detect critical vulnerabilities in major operating systems and browsers.

2The model, developed in partnership with Nvidia, Google, and others, is reserved for select partners to prevent abuse.

3Anthropic is investing $100 million in credits to support the use of Mythos Preview by its partners, including giants like Microsoft and Apple.

💡Why it matters — This initiative could transform how large companies and governments manage cybersecurity, reducing reliance on human intervention.

A Strategic Partnership to Strengthen Cybersecurity

Anthropic recently unveiled an innovative artificial intelligence model as part of a cybersecurity partnership with tech giants such as Nvidia, Google, Amazon Web Services, Apple, and Microsoft. This project, named Project Glasswing, aims to enable large enterprises, and potentially governments, to report vulnerabilities in their systems with minimal human intervention.

Claude Mythos Preview: A Restricted Use Model

The Claude Mythos Preview model is available to Anthropic's launch partners. However, it will not be released publicly due to security concerns. Newton Cheng, head of Anthropic's red team, explained that this model is intended to provide a "leg up" for cybersecurity defenders against adversaries. Partners will use this model to analyze their systems and identify critical vulnerabilities while assisting them in addressing these issues. Access is intentionally restricted to prevent adversaries from using it for malicious purposes.

Coding and Reasoning Skills for Security

Although Claude Mythos Preview has not been specifically trained for cybersecurity, Anthropic emphasizes that its "strong coding and reasoning skills" are behind its advancements in this area. Newton Cheng declined to share specific details about the model's cybersecurity successes, but a blog post from Anthropic revealed that recently, Mythos Preview reported "thousands of high-severity vulnerabilities, including some in every major operating system and web browser." The model identified these vulnerabilities and developed numerous related exploits entirely autonomously, without human intervention.

A Data Leak and Corrective Measures

The existence of Claude Mythos Preview was first revealed last month following a data leak, which Anthropic attributed to human error. Dianne Penn, head of product management at Anthropic, stated that the company is "taking steps to solidify our processes." She clarified that this leak was not related to software vulnerabilities.

Significant Financial Support for Partners

Mythos Preview will be privately available to Glasswing partners, including companies like JPMorgan Chase, Broadcom, Cisco, CrowdStrike, the Linux Foundation, and Palo Alto Networks, as well as about 40 other organizations. Anthropic is committed to subsidizing the cost of using the model with up to $100 million in usage credits, and $4 million in direct donations to the Linux Foundation and the Apache Software Foundation. In the long term, as Anthropic and other AI companies face pressure to turn a profit, the program could evolve into a paid service, provided it proves effective enough for companies to continue using it.

Collaboration with the U.S. Government

Despite a highly publicized clash with the Trump administration, Anthropic has confirmed ongoing "discussions with U.S. government officials" regarding Claude Mythos Preview and its offensive and defensive cybersecurity capabilities. Dianne Penn indicated that the company has "informed high-level U.S. government officials about Mythos and what it can do," and that it remains "committed to working closely with all levels of government." Newton Cheng mentioned that Anthropic is "engaged with" the government, without specifying the exact interlocutors.