What Is the AI Bill of Rights, and Does It Actually Change Anything?

Darren Lage and co-authors describe the White House’s October 2022 Blueprint for an AI Bill of Rights as a non-binding framework built around five principles for automated systems. Lage’s study found limited direct agency uptake in its…

AI-drafted, human-reviewed

The AI Bill of Rights usually refers to the White House’s October 2022 Blueprint for an AI Bill of Rights, which Darren Lage, Riley Pruitt and Jason Ross Arnold describe as a non-binding framework for governing automated systems. It set out five principles, but it did not itself create enforceable rights or general legal duties. Its practical effect appears mixed: Lage and co-authors found limited direct influence on federal agencies in its first year, while a 2024 OMB memo and a working paper by Alex Oesterling and co-authors treat it as a framework that later implementation efforts could build on.

What the AI Bill of Rights is

In U.S. policy discussions, the phrase AI Bill of Rights usually means the Blueprint for an AI Bill of Rights. Lage, Pruitt and Arnold describe that document as an October 2022 white paper from the White House Office of Science and Technology Policy, and they say it was non-binding.

The same paper says the Blueprint addresses automated systems that can meaningfully affect the public’s rights, opportunities, or access to critical resources or services. On that reading, it is a document about protections for people affected by AI systems, not a charter of rights for AI systems themselves. That is a different debate from questions about what human-rights texts say about AI or the distinction between AI welfare and AI rights.

One reason readers ask whether it changed anything is that its title sounds legislative. Lage, Pruitt and Arnold, however, say the Blueprint was advisory rather than binding. So the central question is not whether it became a statute, but whether it changed practice, later policy writing, or how agencies and researchers talk about AI governance.

The five principles in the Blueprint

Lage, Pruitt and Arnold write: “The Blueprint provided a framework for the ethical governance of artificial intelligence systems, organized around five core principles: safety and effectiveness, protection against algorithmic discrimination, data privacy, notice and explanation about AI systems, and human alternatives and fallback.”

Safe and effective systems. Alex Oesterling, Usha Bhalla, Suresh Venkatasubramanian and Himabindu Lakkaraju say this principle calls for systems to be tested before deployment and monitored after deployment.
Algorithmic discrimination protection. Oesterling and co-authors say users should be protected from unjustified differential or harmful treatment on grounds protected by existing law.
Data privacy. Oesterling and co-authors say only minimal and strictly necessary data should be collected, and that permission for collection and use should be sought and respected.
Notice and explanation. Oesterling and co-authors say people should be told when an automated system is being used and given an explanation of how and why it contributed to an outcome. In plain English, explainability means making the system’s role understandable.
Human alternatives, consideration, and fallback. Oesterling and co-authors say people should, when applicable, be able to request a human alternative or human review.

Those headings matter because later arguments about the Blueprint often turn on them. A defender can say the document supplied a clear checklist. A sceptic can reply that a checklist is not the same thing as binding enforcement.

What it is not

Lage, Pruitt and Arnold say the Blueprint was non-binding and did not itself constitute U.S. government policy. On their account, that sharply limits what the document could do on its own.

Later federal guidance, by contrast, can be more specific. In a 24 September 2024 memorandum on federal AI acquisition, OMB Director Shalanda D. Young states: “This memorandum requires agencies to create or update acquisition policies, procedures, and practices to reflect new responsibilities and governance for AI.”

Young also says agencies remain subject to existing AI-related risk-management requirements and that her memorandum does not supersede broader federal policies outside AI-specific scope. Elsewhere in the memorandum, Young says it governs agencies’ own procurement of AI and does not create rights or obligations for contractors or the public.

That comparison suggests a clean distinction. The Blueprint sets out principles. Later memoranda can translate those principles into instructions for agencies. If a reader asks whether the AI Bill of Rights changed the law by itself, the documents here point to no. If the question is whether it helped shape later policy language, the answer is less clear-cut.

What changed after it was issued

Lage, Pruitt and Arnold tested the Blueprint’s early federal effect by reviewing public records across 15 departments. Their paper states: “Only five departments explicitly mentioned the Blueprint, while 12 took steps aligned with one or more of its principles.”

The same authors conclude: “The non-binding Blueprint seems to have had minimal impact on shaping the U.S. government's approach to ethical AI governance in its first year.”

They also offer an explanation for that result: “Factors like public concerns after high-profile AI releases and obligations to follow direct executive orders likely carried more influence over federal agencies.”

That is the strongest evidence in these documents for the claim that the Blueprint did not directly move much federal behaviour in its first year. Lage, Pruitt and Arnold add that more rigorous study would still be needed to measure broader effects with confidence.

At the same time, the later record is not nothing. Young’s 2024 OMB memorandum says more specific advice and best practices for managing AI risks are available in NIST’s AI Risk Management Framework and the Blueprint for an AI Bill of Rights. Oesterling and co-authors likewise write that the White House executive order on safe, secure and trustworthy AI, and associated OMB guidance, built on the Blueprint.

Oesterling and co-authors then try to make the Blueprint usable. Their working paper walks through how a practitioner might operationalise safety, fairness, privacy, explanation and human fallback, while also noting research gaps and trade-offs. That does not show the Blueprint had direct legal force. It does show the framework remained active as a reference point in later implementation work. Readers who want the broader sequence can see our timeline of the AI rights debate.

The case that it mattered — and the case that it did not

The case that it mattered

The strongest case for saying the Blueprint changed something rests on agenda-setting rather than coercion. Lage, Pruitt and Arnold say it organised AI governance around five principles. Oesterling and co-authors treat those principles as practical design and auditing questions, not just slogans. On that reading, the Blueprint supplied a common vocabulary for safety, discrimination, privacy, explanation and human fallback.

Young’s memorandum supports part of that reading. It does not present the Blueprint as a dead letter. Instead, Young points agencies back to it alongside NIST guidance while imposing concrete acquisition requirements. One reading is that the Blueprint mattered because later, more operational documents could build on its framework.

The case that it did not

The strongest case on the other side is more direct. Lage, Pruitt and Arnold say the Blueprint was non-binding, and their study found limited evidence of direct influence on agency action in its first year. They also note that much of the observed activity had earlier precedents or other drivers.

Oesterling and co-authors reinforce that sceptical case from another angle. Their paper says practitioners often still lack detailed guidance on how to implement principles such as privacy and explanation, and it identifies gaps between regulatory goals and current research. If later memos and long technical summaries are needed to make the framework operational, one reading is that the Blueprint alone did not change much on the ground.

The narrowest answer is therefore the safest one. The AI Bill of Rights appears to have changed the language and framework of U.S. AI governance more clearly than it changed enforceable rights or duties by itself. In the documents here, its practical importance looks largest when later institutions turn its principles into specific procurement rules, testing expectations or implementation guidance.

Frequently asked questions

Is the AI Bill of Rights a law?

No. Lage, Pruitt and Arnold describe the 2022 Blueprint as non-binding. Young’s 2024 OMB memorandum is different because it sets acquisition requirements for agencies, but Young also says that memorandum governs agencies’ procurement of AI and does not create rights or obligations for contractors or the public.

What are the five principles in the AI Bill of Rights?

Lage, Pruitt and Arnold list five principles: safety and effectiveness, protection against algorithmic discrimination, data privacy, notice and explanation, and human alternatives and fallback. Oesterling and co-authors expand on each of those principles in their working paper.

Did U.S. federal agencies follow the Blueprint?

Lage, Pruitt and Arnold report mixed evidence. They write that five departments explicitly mentioned the Blueprint and 12 took steps aligned with at least one of its principles, but they conclude that its first-year direct impact on agency behaviour appears minimal.

Does the AI Bill of Rights have anything to do with rights for AI systems?

Not in the documents used here. Lage, Pruitt and Arnold describe the Blueprint as a framework for automated systems that can affect the American public’s rights, opportunities or access to critical resources or services. Its focus is therefore on people affected by AI systems.

Caveats: The description of the Blueprint’s October 2022 release, White House Office of Science and Technology Policy authorship, and non-binding status relies on Darren Lage, Riley Pruitt and Jason Ross Arnold. The operational detail on implementing each principle relies on the working paper by Alex Oesterling, Usha Bhalla, Suresh Venkatasubramanian and Himabindu Lakkaraju.

Sources

Tags: ai-bill-of-rights · ai-governance · us-policy · automated-systems

← all briefings