Anthropic Unveils Mythos Preview, Forms Glasswing
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic announced Mythos Preview and launched Project Glasswing, a consortium of over 40 tech, cybersecurity, critical‑infrastructure and financial firms—including Microsoft, Apple, Google, AWS, the Linux Foundation, Cisco, Nvidia, and Broadcom—to study AI’s cybersecurity impact.
- Mythos Preview is a new Claude model trained for code that, as a side effect, can discover vulnerabilities, develop exploits, conduct penetration testing, assess endpoints, hunt misconfigurations, and evaluate binaries without source code.
- Logan Graham (Anthropic’s frontier red‑team lead) says Mythos Preview can achieve tasks comparable to a senior security researcher and has already uncovered thousands of critical vulnerabilities, including decades‑old bugs missed by prior reviews.
- Google’s VP of security engineering Heather Adkins welcomed the initiative, stating that AI creates new challenges and opportunities for cyber defense.
- Microsoft’s global CISO Igor Tsyganskiy said joining Project Glasswing gives early access to Claude Mythos Preview to identify and mitigate risk early and augment security solutions for customers.
- Anthropic plans a staggered release of Mythos Preview, beginning with the industry collaboration phase, following coordinated‑vulnerability‑disclosure practices to give developers time to patch before public discussion.
- Project Glasswing aims to expand beyond the initial participants, as Graham warns that a limited rollout could accelerate attackers rather than protect.
Why it matters: By giving Microsoft, Apple, Google, AWS and other leading firms private access to a model that can both find and craft sophisticated exploits, Project Glasswing lets them patch vulnerabilities before they become public, reducing the attack surface for their customers while limiting the window for malicious actors to weaponize the same AI capabilities.
