---
title: "Anthropic Expands Claude Mythos Preview to 200 Partners Before Rivals Ship Without Guardrails"
subtitle: "Anthropic quadruples Project Glasswing access, giving 150 critical infrastructure firms in 15+ countries a model too dangerous for public release."
url: "https://alphasignal.ai/news/anthropic-expands-claude-mythos-preview-to-200-partners-before-rivals-ship"
author: "AlphaSignal Newsroom"
author_url: "https://alphasignal.ai/editorial-team"
company: "Anthropic"
company_url: "https://alphasignal.ai/company/anthropicai"
published: "2026-06-02T13:05:25.000Z"
updated: "2026-06-04T22:54:04.217Z"
type: "News"
topics:
  - "Business"
  - "Llms"
subtopics:
  - "Long Context"
upvotes: 3894
read_time: "5 min"
source_url: "https://www.anthropic.com/news/expanding-project-glasswing"
comments: 0
---

# Anthropic Expands Claude Mythos Preview to 200 Partners Before Rivals Ship Without Guardrails

> Anthropic quadruples Project Glasswing access, giving 150 critical infrastructure firms in 15+ countries a model too dangerous for public release.

Anthropic just opened the gates a little wider on the AI model it has refused to ship to the public. [Project Glasswing](https://www.anthropic.com/news/expanding-project-glasswing), the company's cybersecurity coalition built around Claude Mythos Preview, is growing from roughly 50 launch partners to around 200, with approximately 150 new organizations across more than 15 countries getting access to a model Anthropic considers too capable to release broadly.

The expansion targets sectors that were thin in the original cohort. The new group covers power, water, healthcare, communications, and hardware, and many of the new partners are vendors whose codebases underpin critical infrastructure systems. Sources told CyberScoop that Rubrik, a cloud security and data management company, is part of the latest round, though Anthropic itself has not published a roster.

## Why a model nobody can buy keeps making news

Claude Mythos Preview is not a product you can call from the API in the usual sense. It is a frontier model Anthropic has deliberately gated because of its offensive cyber capabilities. The original cohort, announced in April, included Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks as launch partners, with access priced at $25 and $125 per million input and output tokens via the Claude API, Amazon Bedrock, Google Cloud's Vertex AI, and Microsoft Foundry. Anthropic committed $100M in model usage credits to cover Project Glasswing participants throughout the research preview.

What the model can actually do is the reason it matters. On the CyberGym benchmark, which tests vulnerability reproduction, Mythos scored 83.1% compared to Claude Opus 4.6's 66.6%. In internal red-teaming, Anthropic surfaced a flaw in OpenBSD that had gone undetected for 27 years, a bug in FFmpeg's codebase that had survived 16 years and five million passes by automated scanners, and a chain of Linux kernel weaknesses that escalated a basic user account to root without human guidance.

## 10,000 critical bugs and a new bottleneck

Two months into the program, the numbers are striking. The initial partners have found more than 10,000 high- or critical-severity security flaws scanning their own codebases. Anthropic's framing of who is being protected is equally blunt: for most partners, a major attack could affect more than 100 million people, with important ramifications for both global and national security.

But surfacing bugs has stopped being the hard part. Progress on software security used to be limited by how quickly we could find new vulnerabilities. Now it's limited by how quickly we can verify, disclose, and patch the large numbers of vulnerabilities found by AI. That is reshaping how partners use the model. Many of Project Glasswing's partners now use Mythos Preview to write patches and for pre-release checks that prevent vulnerabilities from appearing in the first place. The model is also being used for penetration testing, automating threat detection and response, and rebuilding legacy codebases in memory-safe languages.

## What you can actually touch

If you are not running power grids or browser kernels, Mythos itself is out of reach. But Anthropic has pushed two pieces of the program outward:

- [Claude Security](https://claude.com/product/claude-security), a product that uses Anthropic's latest public frontier models, like Claude Opus 4.8, to scan codebases and suggest patches. It has been used to patch more than 2,100 vulnerabilities in three weeks.
- The internal tooling Glasswing partners use, released on request to trusted security teams. That release includes the skills (custom instructions for repeated work) that partners have built and shared, a harness that helps Claude map the codebase, spin up scanning subagents, triage findings, and write reports, and a threat model builder that maps a codebase to identify potential targets and prioritizes the model's work accordingly.

Cisco, one of the Project Glasswing partners, has also recently open-sourced its Foundry Security Spec to help other defenders build a similar evaluation system.

## The 6-to-12 month clock

The expansion is less a victory lap than a hedge against the calendar. Anthropic expects that within 6 to 12 months many other AI companies will have Mythos-class models, and they could release them without safeguards that prevent misuse. In that world, cyberattacks could occur much more often, and in much more unpredictable forms.

Competitive pressure is already visible. Since Mythos launched, rival OpenAI released its own cybersecurity-focused model, GPT-5.5-Cyber, which it has rolled out to a large group of partners for testing. Policy is moving too: the Trump administration signed a scaled-back executive order on AI security hours after Anthropic's announcement, setting up a voluntary framework requiring AI developers to submit advanced models to a government review up to 30 days before public release.

Anthropic's own roadmap leans into that pressure. It plans to further scale its Cyber Verification Program, which would grant Mythos-class capabilities to additional organizations for specific defensive tasks, with the long-term goal of shifting AI's role in cybersecurity from finding vulnerabilities to a full-cycle model covering disclosure, patching, and deployment. The pitch is a permanent defender advantage. The risk is that someone else ships a comparable model first, with fewer guardrails, and turns the same capability loose on everyone whose code is not yet inside the coalition.

## Sources

- [X post by @anthropicai](https://x.com/anthropicai/status/2061796327986454883)
- [Expanding Project Glasswing](https://www.anthropic.com/news/expanding-project-glasswing)
- [Project Glasswing: An initial update](https://www.anthropic.com/research/glasswing-initial-update)
- [Project Glasswing](https://www.anthropic.com/glasswing)
- [Anthropic expanding access to Project Glasswing](https://cyberscoop.com/anthropic-project-glasswing-expansion-critical-infrastructure-claude-mythos/)
- [Anthropic scales Claude Mythos to critical infrastructure in 15+ countries](https://techcrunch.com/2026/06/02/anthropic-scales-claude-mythos-to-critical-infrastructure-in-15-countries/)