Who decides when a cyber AI tool is safe to deploy? (www.techradar.com)

OpenAI and Anthropic are at an impasse regarding the sharing of their advanced AI cybersecurity tools with European regulators. OpenAI has offered access to its model, while Anthropic is hesitant, citing differing stages in discussions. This debate highlights an essential policy question within the AI/ML community: what is the right balance between openness and restriction in the deployment of powerful AI tools? Recent advancements, like Anthropic's Mythos—which autonomously executed a 32-step simulated corporate cyber attack—underscore the urgency for regulatory frameworks, especially as organizations that will face these AI-enhanced threats are currently ill-equipped to defend against them. According to a recent survey, only 27% of UK organizations feel prepared for AI-driven cyberattacks, with the majority lacking sufficient readiness despite an awareness of increased risks. The evolving nature of AI means that cybersecurity cannot rely solely on tools or software; it requires skilled professionals who can make crucial judgments quickly. Investing in training and certification for cybersecurity teams is shown to significantly reduce cyber risk, making it a critical focus for organizations as regulatory standards around AI security continue to develop. As AI tools grow more sophisticated and adaptive, the gap between current readiness and necessary capabilities must be bridged to ensure robust defenses against increasingly complex threats.