A group of more than 300 scientists from 32 countries has called for urgent changes to AI development and deployment, warning that failures in cybersecurity could damage critical infrastructure. They recommend developing and testing advanced AI systems in tightly controlled environments, including “air-gapped” systems. This call comes as Anthropic warned that its AI models could soon create self-spreading malware affecting internet-connected devices, and industry executives are privately preparing for catastrophic AI-related incidents that could disrupt financial services and critical infrastructure.