Anthropic Launches Project Glasswing to Give AI Cybersecurity Access to Top Tech Firms
Anthropic launched Project Glasswing in April 2026, giving select companies including CrowdStrike, Microsoft, Apple, and Google access to a preview of its Claude Mythos2 model for defensive cybersecurity work. The model has reportedly uncovered thousands of severe vulnerabilities in major operating systems and browsers.
Anthropic launched Project Glasswing in April 2026, giving a select group of technology companies access to a preview of its Claude Mythos2 model for defensive cybersecurity work. The initial partners include CrowdStrike, Microsoft, Apple, and Google.
The model has already produced results. According to The Wall Street Journal, it has uncovered thousands of severe vulnerabilities in major operating systems and browsers. Anthropic said the work demonstrates AI's potential to change the speed and scale of software defense.
The launch comes as AI companies face rising pressure to show they can help protect critical systems, even as their own models raise concerns about potential offensive misuse. OpenAI is also developing advanced cybersecurity capabilities and plans to release them to a limited set of partners.
For cybersecurity companies, the development cuts both ways. AI tools could automate painful but essential work: code review, bug hunting, and threat triage. They could also compress margins for firms whose advantage depends on labor-intensive services.
The broader context is significant. Attackers are moving faster. A critical vulnerability in the open-source Python notebook tool Marimo was exploited within nine hours of public disclosure in April 2026. Iranian threat actors conducted a widespread password-spraying operation against more than 300 Israeli organizations via Microsoft 365. North Korea-linked actors deployed over 1,700 malicious packages impersonating legitimate developer tools.
Google is also rolling out Device Bound Session Credentials in Chrome 146 on Windows, with macOS support to follow. The feature binds session cookies to the user's device, making stolen cookies much less useful to attackers.
Cisco unveiled a Zero Trust architecture for autonomous AI agents. Linx Security raised $50 million for AI-powered identity security systems.
The cybersecurity sector is shifting. AI is moving from assistant to active participant in both offense and defense.
