Meta has released a policy document outlining the circumstances under which the company may choose not to launch certain categories of artificial intelligence systems considered 'risky'.