Anthropic launches Project Glasswing, Claude Mythos Preview in AI cybersecurity push

Anthropic announced Project Glasswing, an effort that aims to use the Claude Mythos Preview model to find and fix software vulnerabilities. The model won't be made generally available but will be shared with more than 40 cybersecurity players to secure AI infrastructure.

The cybersecurity project lands as Anthropic has found that Claude models are better than humans at finding and fixing vulnerabilities. Anthropic noted that Mythos has found vulnerabilities in every browser and operating system.

In a blog post, Anthropic argued that releasing Mythos broadly would likely result in a cybersecurity nightmare since it could be exploited. Anthropic's initial Project Glasswing partners include Amazon Web Services, Anthropic, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, Nvidia and Palo Alto Networks.

Anthropic said:

"Mythos Preview has already found thousands of high-severity vulnerabilities, including some in every major operating system and web browser. Given the rate of AI progress, it will not be long before such capabilities proliferate, potentially beyond actors who are committed to deploying them safely. The fallout—for economies, public safety, and national security—could be severe. Project Glasswing is an urgent attempt to put these capabilities to work for defensive purposes."

The launch partners for Project Glasswing will use Mythos Preview to build defenses. Anthropic said it will commit up to $100 million in usage credits for Mythos Preview usage and $4 million in direct donations to open-source security groups.

Anthropic said Project Glasswing is a start. "The work of defending the world’s cyber infrastructure might take years; frontier AI capabilities are likely to advance substantially over just the next few months. For cyber defenders to come out ahead, we need to act now," said Anthropic.

A few thoughts:

• Anthropic has a point about the cybersecurity issues with LLM releases and frontier models.
• The cadence for new frontier models could be to share with partners, secure infrastructure and then release to the public.
• Anthropic's Project Glasswing puts the company in the middle of the cybersecurity industry. The company has a team focused on everything from finding vulnerabilities to responsible disclosure.
• It's clear that something will have to be done as frontier models advance. There are vulnerabilities that have been in systems for years and just haven't been exploited yet. Anthropic noted that Claude Mythos Preview has found flaws that have "survived decades of human review."
• Project Glasswing is not only a good approach for the cybersecurity industry and Anthropic but also is great marketing for the Claude family of models.

More Anthropic:

• Anthropic inks deal with Google, Broadcom for TPU capacity
• Anthropic vs. SaaS: A Nuanced View