That's what the legal department at my employer (huge multinational corp) came up with: When an employee uses one of the approved gen-ai models it's on the employee to check the output isn't infringing - legal argues that not doing so would be grossly negligent, making the employee personally liable.