Anthropic opens its most powerful AI models to more security teams (www.reuters.com)

Anthropic has announced an expansion of its Cyber Verification Program (CVP), allowing vetted cybersecurity teams greater access to its advanced AI models with reduced safeguards. This decision builds on the success of its Project Glasswing initiative, which detected over 129,000 verified software vulnerabilities from April to July alone. Anthropic's ability to identify and address these vulnerabilities, including an additional 5,500 confirmed through their own scanning, highlights the potential of AI to enhance cybersecurity efforts significantly. The revamped CVP introduces three distinct tiers of access, each tailored for various security needs, from incident response to specialized system testing. Notably, the highest tier allows select organizations to use the AI models on safety-critical systems such as power grids and financial infrastructures, ensuring that powerful AI tools are utilized under stringent oversight. This initiative is particularly crucial as concerns grow around the ability of AI to potentially exploit software vulnerabilities, emphasizing the dual-use nature of such technologies and the necessity of robust security measures in the AI/ML community.