BT Group has joined Anthropic’s Project Glasswing, marking a significant step in enhancing cybersecurity within UK telecommunications. This initiative employs advanced AI to identify vulnerabilities in critical infrastructure, positioning BT as the first UK company to participate in such a program.
The announcement occurred during the UK Government’s AI Adoption Summit, emphasizing BT’s commitment to reinforcing cybersecurity across telecommunications in the country. By utilizing Anthropic’s Claude Mythos Preview model, BT aims to automatically scan extensive codebases for security defects, far surpassing the capabilities of conventional vulnerability assessments.
Since launching in early April 2026, Project Glasswing has detected over 10,000 severe vulnerabilities across the codebases of its partner organizations. This figure highlights the model’s ability to process and analyze code at an unrivaled pace, making it a pivotal asset in identifying potential exploits that might compromise systems, data integrity, or service continuity.
It is crucial to understand the stringent access controls surrounding the Claude Mythos model. While it serves as a powerful tool for locating defensive weaknesses, it also poses risks if used for malicious purposes. Consequently, Anthropic is careful with who can leverage its capabilities.
Initially, Project Glasswing showcased around 50 major organizations, including tech giants like AWS, Apple, and Google. By June 2026, participation surged to approximately 200 organizations across more than 15 countries. BT’s involvement is particularly significant, representing a proactive approach to AI-driven security measures within the UK’s telecommunications sector, and underlining the national importance of securing digital infrastructure.
The alarming number of vulnerabilities discovered suggests that Project Glasswing is fundamentally altering how companies approach security. Organizations involved in this initiative gain a vital advantage, utilizing AI for vulnerability detection that fosters more robust defenses compared to their competitors relying on outdated methods.
However, this concentration of vulnerability data poses its own security issues. If compromised, attackers could exploit these insights to identify critical weaknesses in key infrastructures. Thus, Anthropic’s careful access management is essential for protecting the integrity of the findings from this AI-driven initiative.