Anthropic's Claude artificial intelligence models restrict users from applying automated vulnerability patches during active cyberattacks