Anthropic expanded its Project Glasswing initiative on June 2, 2026, granting access to its Claude Mythos AI model to nearly 200 organizations worldwide [1].
This expansion marks a significant shift in how AI developers collaborate with external entities to mitigate systemic risks. By sharing a high-capability model with a diverse set of partners, the company aims to identify vulnerabilities before they can be exploited by malicious actors.
The initiative grew from an initial group of about 50 organizations [2]. Anthropic recently added access for 150 more organizations [2] to broaden the scope of its safety testing and collaborative research.
Participating organizations include global technology leaders such as Microsoft, Nvidia, and Google [3]. The program also extends to Indian critical-infrastructure operators [3] to ensure that the AI's deployment remains secure across different geopolitical and industrial contexts.
According to the company, the goal of the expansion is to strengthen collaboration on AI safety and address growing concerns regarding AI-driven cyber-risks [1]. The company said the program is designed to promote the responsible development of technology by allowing a wider array of experts to stress-test the model's boundaries.
This effort comes as the industry faces increasing pressure to prove that frontier models will not facilitate large-scale cyberattacks. By integrating critical-infrastructure operators into the feedback loop, the company can better understand the real-world implications of the Claude Mythos model's capabilities [3].
“Anthropic expanded its Project Glasswing initiative... granting access to its Claude Mythos AI model to nearly 200 organisations worldwide”
The expansion of Project Glasswing suggests a transition from internal safety red-teaming to a more open, ecosystem-based approach to AI security. By involving both big tech competitors and national infrastructure operators, Anthropic is attempting to create a standardized safety baseline that transcends individual corporate interests to protect global digital stability.
