Both OpenAI and Anthropic are restricting access to their most powerful cybersecurity capabilities, releasing them only to vetted organizations.