The Blueprint for an AI Bill of Rights is a 2022 framework of five principles from the White House Office of Science and Technology Policy for designing and using automated systems. It is guidance, not a statute, and its pages are archived, so treat it as a reference for good practice.
What is it, and what is its status?
OSTP describes the Blueprint as a set of five principles and associated practices to help guide the design, use and deployment of automated systems to protect the rights of the American public. It calls the framework a national values statement and toolkit that is sector agnostic, and it comes with a handbook called From Principles to Practice.
On status, there are two things to know. First, the OSTP site carrying it is labelled historical material frozen in time, no longer updated. Second, the Blueprint's own page on its relationship to law says some of the protections are already required by the Constitution or existing laws, while ensuring some of the additional protections would require new laws or new policies and practices. In other words, it is guidance that sits next to law without replacing it. Check current federal and state requirements separately.
What are the five principles?
- Safe and effective systems. You should be protected from unsafe or ineffective systems. The text calls for consultation with diverse communities, pre-deployment testing, risk identification and mitigation, and ongoing monitoring.
- Algorithmic discrimination protections. You should not face discrimination by algorithms, and systems should be designed and used equitably. The document defines algorithmic discrimination as automated systems contributing to unjustified different treatment or impacts that disfavor people on protected characteristics.
- Data privacy. You should be protected from abusive data practices through built in protections and have agency over how data about you is used. Collection should conform to reasonable expectations and be limited to data strictly necessary for the context.
- Notice and explanation. You should know when an automated system is in use and understand how and why it contributes to outcomes affecting you, through plain language documentation.
- Human alternatives, consideration and fallback. Where appropriate you should be able to opt out in favor of a human alternative and reach a person who can quickly consider and remedy problems.
What counts as an automated system?
The definitions page says automated systems include systems derived from machine learning, statistics or other data processing or AI techniques, and exclude passive computing infrastructure, meaning intermediary technology that does not influence or determine the outcome of a decision. So a rules engine or a statistical score can fall within scope, not only a large model.
How do the principles become engineering work?
Constructed example: a lender uses a scoring model to pre-screen applications.
- Safe and effective: test the model on data resembling real applicants before launch, set monitoring for drift, and name who can pause it.
- Discrimination: compare outcomes across relevant groups where you lawfully can, and record what you found and what you changed.
- Data privacy: remove input fields that are not needed for the decision, and document where each field comes from.
- Notice and explanation: tell applicants a model contributed, and give a plain reason code a person can act on.
- Human fallback: provide an appeal route to a person who has authority to change the outcome, with a stated response time.
None of this requires the framework to be law. Each item is also a sensible control for audit and risk.
How does it relate to other frameworks?
It is shorter and more rights oriented than risk management frameworks such as the NIST AI Risk Management Framework, which is a separate document. Teams commonly use the Blueprint's principles for goals and a formal framework for process. Treat any mapping as your own judgment and document it.
How to check or apply this
- Inventory automated systems that influence decisions about people, including simple scoring rules.
- For each, answer five questions, one per principle, in a short document.
- Add a pre-launch test and a monitoring plan for each system.
- Write a plain language notice and a human appeal path for decisions with significant effects.
- Cut data fields you cannot justify.
- Check applicable laws separately, since the Blueprint does not replace them.
Quick check
Is the AI Bill of Rights legally binding?
No. It is a framework of principles and practices. Some of its themes overlap with existing law, but the document itself says further protections would need new laws or policies.
Does it only apply to large AI models?
No. The definition covers systems derived from machine learning, statistics or other data processing that influence decisions, and excludes passive infrastructure.
Sources
- OSTP, What is the Blueprint for an AI Bill of Rights?
- OSTP, Relationship to Existing Law and Policy
- OSTP, Blueprint for an AI Bill of Rights (principles)
- OSTP, Definitions
Last reviewed: 18 October 2026.
