Anthropic has officially unveiled a preview of Mythos, its most advanced frontier AI model to date, specifically designed to spearhead a new cybersecurity initiative. The launch follows a previously leaked memo that identified the model as a massive leap in capability for the AI startup.
Project Glasswing: A Defensive Security Shield
The release of Mythos is the cornerstone of “Project Glasswing,” a specialized security program involving 12 key partner organizations. These partners are utilizing the model to fortify critical software systems through defensive security measures. Although Mythos was not built exclusively for cybersecurity, its sophisticated agentic coding and reasoning skills make it highly effective at scanning both first-party and open-source software for vulnerabilities.
Anthropic reports that during initial testing over the past few weeks, Mythos successfully identified thousands of zero-day vulnerabilities, many of which had remained undetected for one to two decades.
Industry Collaboration and Strategic Access
The elite group of partners participating in the preview includes major tech titans such as Apple, Amazon, Microsoft, Cisco, Broadcom, CrowdStrike, the Linux Foundation, and Palo Alto Networks. These organizations are tasked with sharing insights derived from the model’s performance to benefit the broader technology ecosystem. While the preview is restricted, access has been extended to a total of 40 organizations.
Anthropic’s frontier models, known for being their most sophisticated systems, are engineered for high-complexity tasks. Despite this, the company remains in a complex position regarding federal collaboration, currently embroiled in a legal battle with the Trump administration over Pentagon supply-chain risk designations.
From Data Leak to Security Asset
The existence of Mythos—originally codenamed “Capybara”—was first exposed in a major data security incident reported by Fortune last month. The leak, which Anthropic attributed to human error, revealed an unsecured cache of documents that described the model as significantly larger and more intelligent than the existing Opus series.
The internal documentation highlighted the dual-use nature of the technology, noting that while Mythos is being deployed for remediation, its advanced ability to uncover bugs could pose a significant cybersecurity threat if misused by malicious actors. This development follows a separate incident last month where Anthropic accidentally exposed nearly 2,000 source code files during a software update, leading to a massive cleanup effort across GitHub repositories.
Correction April 7, 2026: An earlier version of this article erroneously stated how many partners are working with Anthropic on Project Glasswing. There are 12 partner organizations, though 40 organizations total will have access to the Mythos preview.