Anthropic and OpenAI spark new race for frontier AI access

AI labs are adopting selective trusted-access programs to control powerful cyber-capable models while commercializing them, shifting cybersecurity advantage toward model access. Anthropic is releasing Fable 5 with guardrails and planning a trusted-access program for less restricted Mythos 5, while OpenAI already grants vetted defenders access to models with fewer guardrails.