Project Glasswing
Japanese プロジェクト・グラスウィング（Project Glasswing）
Project Glasswing is a security effort Anthropic started on April 7, 2026. It gives an AI model that is not public to chosen companies and groups, so they can find and fix weaknesses in critical software before attackers do.
Project Glasswing is a plan announced by Anthropic. It keeps an AI model that is very good at finding software weaknesses, Claude Mythos Preview, away from the public and gives it only to 12 launch partners, including Amazon Web Services, Apple, Google and Microsoft, plus more than 40 other organizations that build or maintain critical software. It is like letting building managers check the locks before a thief learns how keys are cut.
Watch out This follows Anthropic's own announcement, and how much has been independently verified is not clear. Anthropic says partners use the model to defend their own and open-source systems. A small company's website does not join directly, so the practical step for an owner is keeping old PHP and plugins updated.
Source: Project Glasswing: Securing critical software for the AI era · Expanding Project Glasswing (checked 2026-10-26)