Instead of releasing the model broadly, Anthropic restricted access to a small group of vetted partners across technology and cybersecurity. Participants include major companies such as Amazon Web Services, Apple, Google, Microsoft, Nvidia, CrowdStrike, and Palo Alto Networks.

The goal is to test how powerful AI systems can help protect software systems that billions of people depend on—from cloud infrastructure to enterprise platforms—before those capabilities become widely available.

What Claude Mythos Preview Is Designed to Do

Claude Mythos Preview is described as Anthropic’s most advanced unreleased AI model. While it is a general‑purpose model, its strong coding and reasoning abilities make it particularly effective at analyzing software systems and identifying vulnerabilities.

In Project Glasswing, partners use the model for defensive cybersecurity tasks, including:

• Scanning first‑party and open‑source code for security flaws
• Identifying previously overlooked vulnerabilities
• Analyzing potential cyberattack paths
• Sharing intelligence about discovered threats with other affected organizations

Anthropic initially limited how security findings could be shared, but later allowed participants to distribute threat information to organizations that might face similar vulnerabilities—encouraging faster defensive coordination.

Why IBM Joined the Initiative

IBM announced its participation in Project Glasswing as part of an expansion of its AI‑powered enterprise security portfolio. The company framed the move as preparation for an era in which cyber adversaries increasingly use advanced AI tools.

From IBM’s perspective, models like Claude Mythos are significant because they can reveal weaknesses in software systems that traditional security methods might miss. According to IBM’s analysis of the technology, AI systems with strong coding and reasoning abilities can expose vulnerabilities that have remained hidden in complex systems for years.

Joining the program gives IBM early visibility into these capabilities and allows it to integrate lessons from the research into its enterprise security offerings.

How AI Is Changing the Cybersecurity Landscape

Anthropic launched Project Glasswing partly because of concerns about how rapidly advancing AI could affect cybersecurity. The company has warned that frontier AI models may increase the likelihood of large‑scale AI‑enabled cyberattacks if their capabilities are widely available without safeguards.

By giving trusted partners early access, the initiative aims to ensure defenders understand the technology before attackers do.

The approach reflects a broader trend in cybersecurity strategy: using advanced AI to detect vulnerabilities faster than human researchers alone could manage. If effective, tools like Claude Mythos could dramatically accelerate vulnerability discovery and threat analysis across large software ecosystems.

Where This Fits in IBM’s Enterprise Security Strategy

IBM’s security business focuses heavily on protecting large organizations running complex digital infrastructure—particularly in sectors such as finance, telecommunications, and healthcare. These industries depend on large, interconnected software systems where vulnerabilities can have systemic consequences.

By participating in Project Glasswing, IBM gains access to research and tooling that could strengthen:

• enterprise vulnerability discovery
• threat intelligence analysis
• defensive coordination across organizations

The partnership is part of IBM’s broader effort to equip clients with security tools capable of responding to attackers who are increasingly using AI themselves.

Why This Partnership Matters

Project Glasswing signals a new phase in the AI‑cybersecurity relationship. Instead of simply reacting to threats, technology companies are experimenting with frontier AI models as defensive tools—testing whether they can identify risks earlier and protect critical systems at scale.

IBM’s involvement shows how seriously major enterprise vendors are taking that possibility. As AI systems become more capable, the cybersecurity race may increasingly come down to who can deploy those capabilities for defense first.

Instead of releasing the model broadly, Anthropic restricted access to a small group of vetted partners across technology and cybersecurity. Participants include major companies such as Amazon Web Services, Apple, Google, Microsoft, Nvidia, CrowdStrike, and Palo Alto Networks.

The goal is to test how powerful AI systems can help protect software systems that billions of people depend on—from cloud infrastructure to enterprise platforms—before those capabilities become widely available.

What Claude Mythos Preview Is Designed to Do

Claude Mythos Preview is described as Anthropic’s most advanced unreleased AI model. While it is a general‑purpose model, its strong coding and reasoning abilities make it particularly effective at analyzing software systems and identifying vulnerabilities.

In Project Glasswing, partners use the model for defensive cybersecurity tasks, including:

• Scanning first‑party and open‑source code for security flaws
• Identifying previously overlooked vulnerabilities
• Analyzing potential cyberattack paths
• Sharing intelligence about discovered threats with other affected organizations

Anthropic initially limited how security findings could be shared, but later allowed participants to distribute threat information to organizations that might face similar vulnerabilities—encouraging faster defensive coordination.

Why IBM Joined the Initiative

IBM announced its participation in Project Glasswing as part of an expansion of its AI‑powered enterprise security portfolio. The company framed the move as preparation for an era in which cyber adversaries increasingly use advanced AI tools.

From IBM’s perspective, models like Claude Mythos are significant because they can reveal weaknesses in software systems that traditional security methods might miss. According to IBM’s analysis of the technology, AI systems with strong coding and reasoning abilities can expose vulnerabilities that have remained hidden in complex systems for years.

Joining the program gives IBM early visibility into these capabilities and allows it to integrate lessons from the research into its enterprise security offerings.

How AI Is Changing the Cybersecurity Landscape

Anthropic launched Project Glasswing partly because of concerns about how rapidly advancing AI could affect cybersecurity. The company has warned that frontier AI models may increase the likelihood of large‑scale AI‑enabled cyberattacks if their capabilities are widely available without safeguards.

By giving trusted partners early access, the initiative aims to ensure defenders understand the technology before attackers do.

The approach reflects a broader trend in cybersecurity strategy: using advanced AI to detect vulnerabilities faster than human researchers alone could manage. If effective, tools like Claude Mythos could dramatically accelerate vulnerability discovery and threat analysis across large software ecosystems.

Where This Fits in IBM’s Enterprise Security Strategy

IBM’s security business focuses heavily on protecting large organizations running complex digital infrastructure—particularly in sectors such as finance, telecommunications, and healthcare. These industries depend on large, interconnected software systems where vulnerabilities can have systemic consequences.

By participating in Project Glasswing, IBM gains access to research and tooling that could strengthen:

• enterprise vulnerability discovery
• threat intelligence analysis
• defensive coordination across organizations

The partnership is part of IBM’s broader effort to equip clients with security tools capable of responding to attackers who are increasingly using AI themselves.

Why This Partnership Matters

Project Glasswing signals a new phase in the AI‑cybersecurity relationship. Instead of simply reacting to threats, technology companies are experimenting with frontier AI models as defensive tools—testing whether they can identify risks earlier and protect critical systems at scale.

IBM’s involvement shows how seriously major enterprise vendors are taking that possibility. As AI systems become more capable, the cybersecurity race may increasingly come down to who can deploy those capabilities for defense first.