UNESCO Recommendation on the Ethics of AI PA1a: paras 48 to 52 Ethical impact assessment and due diligence: identify benefits and risks, test high-risk systems before release, and monitor the whole life cycle

Member States should put effective measures in place and ensure other stakeholders follow them, encouraging tools for assessing effects on human rights, the rule of law and democracy, and for due diligence, consistent with the UN Guiding Principles on Business and Human Rights. Impact assessment frameworks should identify benefits, concerns and risks and prevention, mitigation and monitoring measures, covering human rights (especially of vulnerable people), labour rights, the environment and social effects, with citizen participation. Member States and companies should develop due diligence and oversight mechanisms, assess effects on poverty and the digital divide, implement enforceable transparency protocols, study effects of AI recommendations on decision-making autonomy, test systems that may threaten human rights widely (in real-world conditions if needed) before release, and monitor algorithms, data and actors across the life cycle.