Human rights in the age of

HUMAN RIGHTS IN THE AGE OF ARTIFICIAL INTELLIGENCE
3. Be transparent about efforts to identify, prevent, and mitigate the harms in AI systems. Transparency to 
all individuals and groups impacted as well as other relevant stakeholders is a key part of human rights due 
diligence, and involves communication.
123
In practice, this means that private-sector actors must:
• Publicly disclose information on identified human rights risks, including both how the system is 
designed or the context in which it is used.
• Publish technical details about the AI system, including samples of training data and details about 
sources of the data.
2. Provide transparency and explainability to the extent possible. 
Private-sector actors must be highly 
transparent and provide meaningful information about how AI systems work. Transparency is especially 
important when AI systems may have a significant public or personal impact, for example in medicine or in 
content recommendation and moderation. Specifically, this includes:
• Adherence to open source and open data standards.
• Publication of meaningful, accessible explanations of how an AI system works so that people can be 
meaningfully informed about how it may impact them. 
3. Establish appropriate mechanisms for accountability and remedy. 
Private-sector actors should 
establish internal accountability mechanisms for the functioning of AI systems. Although states have the 
primary duty to provide access to formal remedy in the case of human rights violations, companies must take 
additional action to ensure access to meaningful, effective non-judicial remedy and redress.
124
At minimum, 
this includes:
• Internal responsibility for development and implementation of an AI system.
• Commitment by third parties developing AI systems for third parties to clearly delineating 
responsibility and accountability between vendor and client, including the vendor’s obligation to ensure 
proper training of the risks of the system as well as to mitigate the risk of function creep and misuse 
of an AI system. 
• Creation of clear, transparent processes by which an individual can directly submit complaints and 
seek redress for human rights harms in a timely manner. These could be administered internally, 
in collaboration with other relevant stakeholders, or through a mutually acceptable external body.
125
Findings should feed back into product and policy development to better prevent and mitigate harms.