Anthropic, Amazon, Microsoft, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, NVIDIA, and Palo Alto Networks… What do all of these companies have in common besides their massive tech industry presence? Apparently, they all care about AI models able to discover “thousands of high-severity vulnerabilities, including some in every major operating system and web browser”. While Anthropic claims to have one, the main question is who and how will use it? Project Glasswing aims to make sure that such potential serves boosting cybersecurity rather than powering cyberattacks.
On Tuesday, AI company Anthropic, the creator of a popular Claude chatbot, announced its new Project Glasswing initiative. Centered around Anthropic’s testing of Claude Mythos, the frontier model that remains unreleased for now due to its reported stark coding capabilities that make existing security tools fade in comparison, the project unites most influential tech players to secure the global web systems from AI exploitation.
What propelled Project Glasswing launch?
According to the AI developer, Mythos Preview has spotted too many vulnerabilities in existing software powering the global internet. Apparently, this or similar powerful models can not only find but also exploit software vulnerabilities if used by malefactors. Preventing that scenario is one of the main concerns for modern tech industry. Most of this year’s RSA cybersecurity conference in San Francisco was dedicated to this very topic – the rise of AI-powered cyberattacks and how to prevent them when existing tools are not enough anymore.
AI is increasingly becoming a cornerstone in the fight against payment fraud, but, at the same time, it’s increasingly used by fraudsters for their criminal schemes.
“Looking ahead to 2026, AI is going to shape both the threat and the defense, ” predicted Michele Tucci, Chief Strategy Officer and co-founder of Credolab late last year. “Fraudsters will keep using generative and automated tools to scale up attacks and mimic users more convincingly. At the same time, financial institutions will rely on AI models that can read micro-behaviors, detect unusual patterns, and evaluate digital trust in real time.”
As AI tools grow more powerful by the day, it is essential that they are used for right purposes. That’s what brought the Project Glasswing to life. Its official focus is defensive cybersecurity: helping companies and organizations secure their systems using tools that may soon be available to attackers otherwise.
Who participates in Anthropic’s Project Glasswing?
The initiative brought all the industry bright minds under one umbrella. Project Glasswing has gathered a large coalition of major tech and infrastructure players to work with Anthropic’s unreleased model whose capabilities in finding and exploiting vulnerabilities could be misused otherwise. Anthropic is specifically stressing that it shares access only with trusted partners and that the initiative has defensive, ethical character, not being meant for general release. The early partners include:
- Cloud and tech giants: Amazon Web Services, Microsoft, Google, Apple, NVIDIA
- Security and networking firms: CrowdStrike, Palo Alto Networks, Cisco, Broadcom
- Software ecosystem & infrastructure: Linux Foundation
- Financial institution partner: JPMorgan Chase
More than 40 additional organizations are also reportedly getting access to the Mythos preview mode for their cybersecurity work like scanning and securing both first-party and open-source systems. Anthropic is also committing significant funds to this effort, including usage credits for Mythos Preview (about $100 million worth) and donations to open‑source security projects.
Why Project Glasswing matters not just for Anthropic but for the world
The cooperation in this segment is essential for “economies, public safety, and national security” since if the AI potential is not used for proactive defence, world online systems may soon have to battle against the same capabilities used for cyber offences. Whether they’ll win this battle is a good question, considering that Claude Mythos even in its unreleased preview form was able to discover security loopholes that had been undetected for decades.
