Anthropic Expands Claude Access for Vetted Cyber Teams as Glasswing Finds 129,000 Flaws
Summary
Anthropic is expanding a program that allows vetted cybersecurity professionals to test its AI models with fewer restrictions. This initiative, known as Project Glasswing, reportedly identified over 129,000 software vulnerabilities between April and July 2026.
IFF Assessment
FRIEND
This article details efforts to find and fix software vulnerabilities, which is beneficial for cybersecurity defenders.
Defender Context
As AI models become more integrated into critical systems, allowing security professionals to probe them for weaknesses is a crucial step in identifying potential attack vectors. Defenders should stay aware of how AI models are being tested and secured, as vulnerabilities in these systems could lead to new types of exploits.