Anthropic launches Mythos AI for security, 12 partners
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic released a preview of its new frontier model Mythos, described in a leaked memo as one of its most powerful models, for use by a small group of partner organizations in cybersecurity.
- Project Glasswing will involve 12 partner organizations—including Amazon, Apple, Broadcom, Cisco, CrowdStrike, Linux Foundation, Microsoft, and Palo Alto Networks—to deploy Mythos for defensive security work, scanning both first‑party and open‑source software for vulnerabilities.
- Mythos has already identified thousands of zero‑day vulnerabilities, many critical and up to two decades old, according to Anthropic’s claims.
- Anthropic says the model, while not specifically trained for cybersecurity, leverages strong agentic coding and reasoning skills from its Claude AI systems to perform complex tasks such as code analysis.
- Anthropic is holding ongoing discussions with federal officials about Mythos’s use, even as it faces a legal battle with the Trump administration after the Pentagon labeled its lab a supply‑chain risk for refusing autonomous targeting or surveillance of U.S. citizens.
- Capybara was the code name in a leaked draft blog attributed to human error, which claimed the model could be weaponized by bad actors to find and exploit bugs.
- Anthropic previously exposed nearly 2,000 source‑code files and over half a million lines of code in a separate incident, leading to mass GitHub repository takedowns during cleanup.
Why it matters: Tech firms gain early access to an AI that can uncover thousands of critical zero‑day bugs, potentially hardening software supply chains, while Anthropic confronts regulatory scrutiny and a legal battle that could limit its deployment or raise compliance costs for the broader industry.
