Anthropic has announced Project Glasswing, a new cybersecurity initiative built around an unreleased frontier AI model called Claude Mythos Preview. The project brings together a dozen of the largest names in technology and finance — Amazon Web Services, Anthropic, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks — along with more than 40 additional organisations that build or maintain critical software infrastructure.
The goal: use AI’s rapidly advancing offensive coding capabilities for defensive purposes, before those same capabilities fall into the wrong hands.
Anthropic is committing up to $100 million in usage credits for Mythos Preview across the program, plus an additional $4 million in direct donations to open-source security organisations.
Why Project Glasswing Was Formed
Project Glasswing was created in response to capabilities Anthropic observed during testing of Claude Mythos Preview — a general-purpose, unreleased frontier model. According to Anthropic, the model has reached a level of coding ability where it can surpass all but the most skilled humans at finding and exploiting software vulnerabilities.
In internal testing, Mythos Preview has already surfaced thousands of high-severity vulnerabilities, including issues in every major operating system and web browser.
Anthropic’s view is that this kind of capability will not stay rare for long. As frontier AI progress continues, similar tools are likely to spread to actors who may not be committed to safe deployment — with potentially severe fallout for economies, public safety, and national security. Project Glasswing is framed as an urgent attempt to put these capabilities to work on the defensive side first.
What the Launch Partners Will Do
Each of the 12 launch partners will use Claude Mythos Preview as part of their internal defensive security work. Anthropic has said it will share what it learns from these deployments back with the broader industry, so the entire security community benefits — not just the named partners.
Launch Partners
| Category | Partner |
|---|---|
| Frontier AI | Anthropic |
| Cloud | Amazon Web Services, Google, Microsoft |
| Consumer & Devices | Apple |
| Networking & Security | Cisco, Palo Alto Networks |
| Endpoint Security | CrowdStrike |
| Silicon | Broadcom, NVIDIA |
| Financial Services | JPMorgan Chase |
| Open Source | The Linux Foundation |
Beyond the launch group, Anthropic has extended access to more than 40 additional organisations that maintain critical software infrastructure, so they can use Mythos Preview to scan and harden both first-party and open-source systems.
Anthropic’s Financial Commitment
| Commitment | Amount |
|---|---|
| Usage Credits for Mythos Preview | Up to $100 million |
| Direct Donations to Open-Source Security | $4 million |
The donation portion is targeted specifically at open-source security organisations — the maintainers and projects that often defend critical infrastructure on shoestring budgets.
What Claude Mythos Preview Is
Claude Mythos Preview is described as a general-purpose, unreleased frontier model trained by Anthropic. It is not yet available to the public. What makes it relevant to Project Glasswing is its coding capability — specifically, its ability to read source code and identify security flaws at a level Anthropic says exceeds nearly all human reviewers.
Key facts disclosed about the model:
- Status: Unreleased preview, accessible only through Project Glasswing partners
- Type: General-purpose frontier model
- Notable capability: Vulnerability discovery and exploitation in software
- Track record so far: Thousands of high-severity vulnerabilities found, including in every major OS and browser
Anthropic has not yet disclosed when, or whether, Mythos Preview will graduate into a publicly available Claude release.
The Bigger Picture: Defenders Need to Move First
In its announcement, Anthropic was clear that Project Glasswing is a starting point, not a complete solution. The company stated that no single organisation can solve cybersecurity at this scale alone, and called on frontier AI developers, software companies, security researchers, open-source maintainers, and governments to all play a role.
Two timelines are in tension:
- Defending the world’s cyber infrastructure is work that may take years.
- Frontier AI capabilities are likely to advance substantially over just the next few months.
Anthropic’s argument is that for cyber defenders to come out ahead, the work has to start now — before the same class of capabilities is available to adversaries at scale.
Project Glasswing Quick Facts
| Item | Detail |
|---|---|
| Initiative Name | Project Glasswing |
| Lead AI Partner | Anthropic |
| AI Model Used | Claude Mythos Preview (unreleased) |
| Launch Partners | 12 (including AWS, Apple, Google, Microsoft, NVIDIA, JPMorgan Chase, Linux Foundation) |
| Extended Access | 40+ critical infrastructure organisations |
| Anthropic Usage Credit Commitment | Up to $100 million |
| Open-Source Security Donations | $4 million |
| Primary Goal | Use frontier AI to find and fix software vulnerabilities defensively, at scale |
Note: Information in this article is based on Anthropic’s official announcement of Project Glasswing. Specifications and financial commitments listed above may change over the course of the program. Please refer to Anthropic’s official communications for the most up-to-date details.
Disclaimer: This article is informational and reports on a publicly announced initiative. It is not a review or endorsement of any product, service, or company involved.