Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Introduction
Project Glasswing is Anthropic’s $100 million initiative to defend critical infrastructure against AI‑powered cyber attacks. Launched alongside the Anthropic Mythos AI model, Glasswing gives restricted access to Mythos Preview for a select consortium of tech giants, security firms, and financial institutions. The goal: use Mythos’s unprecedented ability to find zero‑day vulnerabilities to patch systems before malicious actors can weaponise similar AI capabilities. This Project Glasswing deep dive covers the participating companies, how the defensive alliance works, the $100 million commitment, and why critics say it is not enough.
For a complete overview of the model itself, read our main guide: Goldman Sachs ‘Hyper‑Aware’ of Anthropic Mythos AI .
What Is Project Glasswing?
Project Glasswing is Anthropic’s response to the dual‑use nature of its Claude Mythos Preview model. Because Mythos can autonomously find and exploit software vulnerabilities, Anthropic decided not to release it publicly. Instead, the company created a controlled access program for defensive purposes.
The name “Glasswing” refers to the transparency needed to counter hidden threats – and also to a species of butterfly with transparent wings, symbolising the invisible nature of cyber defence.
Key components of the initiative:
- Restricted access to Mythos Preview – Only approved partners can use the model.
- $100 million in usage credits – Anthropic is subsidising defensive testing.
- $4 million in direct donations – To open‑source security organisations.
- Collaborative vulnerability disclosure – Partners share findings to patch systems faster.
According to Anthropic’s official announcement, the goal is to “harness frontier AI capabilities to make the digital ecosystem safer for everyone”.
Who Is Participating in Project Glasswing?
Anthropic has assembled a consortium of over 20 organisations, including:
| Category | Participating Organisations |
|---|---|
| Tech giants | Amazon, Apple, Google, Microsoft, NVIDIA, Cisco, Broadcom |
| Security firms | CrowdStrike, Palo Alto Networks, Contrast Security |
| Financial institutions | JPMorgan Chase, Goldman Sachs |
| Open‑source foundations | Linux Foundation, OpenSSF |
| Government partners | UK AI Security Institute (AISI), US CISA (observer) |
These partners receive API access to Mythos Preview and can use it to probe their own systems, audit open‑source software, and identify critical vulnerabilities.
For a deeper look at Goldman Sachs’s involvement, see our Goldman Sachs CEO Warning analysis.
How Does the Defensive Alliance Work?
The Glasswing workflow has four phases:
- Access – Approved partners receive API credentials for Mythos Preview.
- Probing – Partners use Mythos to scan their own codebases, infrastructure, and third‑party dependencies.
- Discovery – Mythos autonomously identifies potential vulnerabilities, including zero‑days.
- Remediation – Partners patch the vulnerabilities and, where possible, disclose them to affected vendors (e.g., open‑source projects).
Anthropic has also committed to sharing anonymised threat intelligence across the consortium. However, participation is voluntary, and some partners may choose to keep their findings private.
For more on the technical capabilities of Mythos, read our UK AISI Mythos Test deep dive.
The $100 Million Commitment – What Does It Fund?
Anthropic is providing $100 million in usage credits for Mythos Preview over the next two years. This covers API calls, compute costs, and dedicated support. In addition, Anthropic is donating $4 million directly to open‑source security organisations such as the Open Source Security Foundation (OpenSSF) and the Linux Foundation.
The credits are distributed based on each partner’s scope and contribution. For example:
- Major tech companies (Google, Microsoft) receive larger credits to audit their extensive codebases.
- Security firms (CrowdStrike, Palo Alto) receive credits to integrate Mythos into their threat intelligence platforms.
- Open‑source foundations receive funding to patch critical projects.
The Fix Gap – Why Finding Vulnerabilities Is Not Enough
Despite the promise of Project Glasswing, some security experts remain sceptical. David Lindner, CISO at Contrast Security, told Fortune: “We’ve never had a problem finding vulnerabilities. We find them every day. We actually have a pile of them that we just don’t fix”.
According to the 2026 Verizon Data Breach Investigations Report, over 60% of known vulnerabilities remain unpatched for more than a year after disclosure. Mythos can find thousands of bugs, but without the resources to fix them, the defensive value is limited.
Anthropic acknowledges this gap. The company is working with partners to prioritise patching for critical vulnerabilities (CVSS score 9.0+) and to automate remediation where possible. Nevertheless, the “fix gap” remains a fundamental challenge.
For a broader discussion of AI’s dual‑use nature, see our AI Cyber Defence vs Offence guide.
Comparison Table – Project Glasswing vs Traditional Bug Bounties
| Aspect | Project Glasswing | Traditional Bug Bounty |
|---|---|---|
| AI‑powered discovery | ✅ (Mythos autonomously finds bugs) | ❌ (Human researchers) |
| Scale | Thousands of vulnerabilities per week | Dozens per week (large programs) |
| Cost | $100M in credits (subsidised) | Varies ($500‑$50,000 per bug) |
| Access restriction | Closed consortium | Open to anyone |
| Remediation focus | Partial (prioritises critical) | Full (if bounty paid) |
| Open‑source support | ✅ ($4M donations) | ❌ (rarely) |
Project Glasswing complements, rather than replaces, traditional bug bounties. Human researchers still excel at complex, logic‑based flaws that AI may miss.
Real‑World Applications of Project Glasswing
- For financial institutions: Banks like Goldman Sachs and JPMorgan Chase use Mythos to test their trading systems and customer portals.
- For cloud providers: Amazon and Google scan their infrastructure for misconfigurations and zero‑days.
- For open‑source projects: The Linux Foundation receives funding and AI‑assisted audits for critical libraries (OpenSSL, OpenSSH, etc.).
- For security vendors: CrowdStrike integrates Mythos into its threat hunting platform to identify novel attack patterns.
FAQ Section
Q1: What is Project Glasswing?
A: It is Anthropic’s $100 million initiative to give restricted access to its powerful Claude Mythos AI model to a consortium of tech, security, and financial companies for defensive cyber purposes.
Q2: Which companies are participating?
A: Participants include Amazon, Apple, Google, Microsoft, NVIDIA, CrowdStrike, JPMorgan Chase, Goldman Sachs, the Linux Foundation, and others.
Q3: Is Mythos being released to the public?
A: No. Mythos is too dangerous for public release. Only approved Glasswing partners have access.
Q4: Does Glasswing solve the problem of unpatched vulnerabilities?
A: Not entirely. Finding vulnerabilities is only half the battle; many organisations lack resources to patch them. Glasswing prioritises critical fixes and supports open‑source projects, but the “fix gap” remains.
Conclusion
Project Glasswing is an unprecedented experiment in defensive AI. By giving restricted access to its most powerful model, Anthropic hopes to flip the asymmetry of cyber conflict – using AI to defend rather than attack. The $100 million commitment and consortium of industry giants signal serious intent. Yet the “fix gap” and the voluntary nature of participation raise questions about whether Glasswing can truly outpace malicious actors. As Mythos finds thousands of vulnerabilities, the real test will be how quickly the world patches them.
Next step: Explore the UK AISI’s independent evaluation of Mythos in our UK AISI Mythos Test deep dive.
