Anthropic Expands Cyber-Focused Claude Access for Approved Defenders
Anthropic has integrated Project Glasswing into a multi-tier access program for its advanced cybersecurity large language models. The program includes models such as Opus, Sonnet, and Mythos and offers fewer guardrails to approved defenders.
Anthropic's move places Project Glasswing within a multi-tier access program for cybersecurity-focused Claude models. Approved defenders are offered advanced models such as Opus, Sonnet, and Mythos with fewer guardrails. This reflects a wider challenge in AI: making powerful tools available for defense while limiting who can use them and under what conditions. The available announcement does not detail approval criteria, oversight, or technical limits.
The shift could affect cybersecurity teams, AI developers, and organizations relying on defensive tools. Approved defenders may gain stronger assistance against threats, while broader access remains limited. If fewer guardrails are available only to approved users, oversight and misuse risks may draw scrutiny. The public could benefit from improved protection, but unequal access and dependence on private AI systems may shape how cyber defense evolves.