Accord says companies will implement 'robust internal controls' for their AI systems as concerns mount over safety.