The government is considering introducing risk categories for AI applications, some of which would then be banned.