Anthropic Expands Claude Access for Vetted Cyber Teams as Glasswing Finds 129,000 Flaws

Anthropic is expanding a program that gives vetted cybersecurity teams access to its advanced AI models with limited safeguards. This move comes as the company's Project Glasswing initiative has identified a large number of software vulnerabilities, with at least 129,000 verified flaws found between April and July 2026. The company also discovered additional vulnerabilities, although the exact number is not specified. The expansion of the program allows cybersecurity professionals to test Anthropic's AI models in a more open environment. This is part of Anthropic's efforts to improve the security of its AI models through external testing. The Project Glasswing initiative has been successful in uncovering a significant number of software vulnerabilities in a short period of time.