Last reviewed
May 18, 2026
Sources
No citations yet
Review status
Needs citations
Revision
v1 · 4,081 words
Add missing citations, update stale details, or suggest a clearer explanation.
The Blueprint for an AI Bill of Rights is a non-binding policy framework released by the White House Office of Science and Technology Policy (OSTP) on October 4, 2022, during the Biden-Harris administration.[^1] Subtitled Making Automated Systems Work for the American People, the seventy-three page document outlines five principles intended to guide the design, development, and deployment of "automated systems" so that they protect civil rights, equal opportunity, and access to critical resources or services for the American public.[^2][^3] The five principles are safe and effective systems, algorithmic discrimination protections, data privacy, notice and explanation, and human alternatives, consideration, and fallback.[^1][^4]
The Blueprint was the culmination of a roughly year-long policy planning process initiated in October 2021 by then-OSTP Director Eric Lander and Deputy Director Alondra Nelson, who jointly authored a WIRED opinion piece calling for "a bill of rights for an AI-powered world."[^5][^6] Nelson, who served as Acting OSTP Director from February to October 2022, led the drafting effort and is generally credited as the principal architect of the framework.[^7] The Blueprint was accompanied by a separate technical companion titled From Principles to Practice, which translated each principle into concrete examples and recommended practices for designers, deployers, and policymakers of automated systems.[^8]
A key feature, repeatedly emphasized in the document itself, is that the Blueprint is a "white paper" that "is non-binding and does not constitute U.S. government policy" and that it "does not supersede, modify, or direct an interpretation of any existing statute, regulation, policy, or international instrument."[^9] This non-binding status drew both praise and criticism. Civil-society and academic commentators welcomed the document as a coherent rights-based framework but criticized the lack of enforcement mechanisms, while parts of industry argued it could stifle innovation and a smaller faction of advocates argued it failed to ban the most harmful uses outright.[^10]
The Blueprint became a foundational reference for U.S. AI policy throughout 2023 and influenced the Biden administration's October 2023 ai executive order (Executive Order 14110), which explicitly built upon it.[^11] It also influenced state legislative efforts, most notably the Colorado Artificial Intelligence Act enacted in May 2024.[^12] Following the January 20, 2025 inauguration of President Donald Trump, the Blueprint was removed from the active whitehouse.gov website and migrated to the Biden White House archive.[^13] Executive Order 14110 was rescinded the same day, and on January 23, 2025 a new executive order, Removing Barriers to American Leadership in Artificial Intelligence, replaced the Biden-era approach. The Trump administration's ai action plan, released on July 23, 2025, formalized a deregulation-focused federal AI agenda that diverged sharply from the Blueprint's rights-based premise.[^14]
The Blueprint originated in the early policy work of the Biden administration's reconstituted OSTP. President Biden had elevated OSTP to a cabinet-level office and named geneticist Eric Lander as its director, with sociologist Alondra Nelson as deputy director for science and society. On October 8, 2021, Lander and Nelson published an opinion column in WIRED titled "Americans Need a Bill of Rights for an AI-Powered World," in which they announced that OSTP would lead a public policy planning process to articulate the rights of Americans in the face of increasingly pervasive data-driven technologies.[^5][^6]
The op-ed gave concrete examples of harms that motivated the effort, including voice-recognition systems that fail for users with certain accents, facial recognition technology associated with wrongful arrests, and clinical risk-stratification algorithms that have under-counted disease severity for some patient populations.[^5] Lander and Nelson floated several possible enforcement mechanisms for a future bill of rights, including the federal government's contracting power, new statutes, and regulatory action by existing agencies.[^5]
Alongside the op-ed, OSTP launched a public engagement process. On October 8, 2021, OSTP published a Notice of Request for Information (RFI) on Public and Private Sector Uses of Biometric Technologies in the Federal Register. The RFI sought input on existing and proposed uses of biometric technologies for identity verification, identification, and inference of attributes such as emotion or mental state. Comments were accepted through January 15, 2022 and OSTP received approximately 130 responses, which became part of the underlying record for the Blueprint.[^15][^16]
In addition to the formal RFI, OSTP conducted panel discussions, listening sessions, and meetings with federal officials, industry representatives, civil society organizations, academics, workers, educators, students, and international partners over the course of 2021 and 2022.[^17][^4] The Blueprint's foreword describes this engagement as "extensive" and notes that contributions from these stakeholders "played a central role in shaping" the document.[^4]
OSTP framed the Blueprint as a continuation of a longer civil-rights tradition. The document's title intentionally invokes the U.S. Bill of Rights, and its foreword situates protections against algorithmic harm in the lineage of civil-rights statutes and constitutional guarantees, while clarifying that the Blueprint itself is not new law.[^4]
Although the Blueprint was issued in the name of the White House and OSTP rather than under any individual byline, Alondra Nelson is widely identified as its principal author and the official who shepherded it through publication.[^7][^18]
Nelson, a sociologist who had previously served as president of the Social Science Research Council and as a professor at Columbia University and the Institute for Advanced Study, joined OSTP in January 2021 as Deputy Director for Science and Society. Following Eric Lander's resignation in February 2022 over reported workplace misconduct findings, Biden named Nelson Acting OSTP Director and Deputy Assistant to the President on February 17, 2022.[^7] In that capacity she was the first Black person and the first woman of color to lead OSTP in the office's 46-year history.[^7] She served as Acting Director until October 3, 2022, the day before the Blueprint was released, when Arati Prabhakar was sworn in as the Senate-confirmed director.[^7]
Other senior OSTP staff who contributed to drafting and policy work included Dr. Sorelle Friedler, Assistant Director for Data and Democracy; Dr. Suresh Venkatasubramanian, Assistant Director for Science and Justice (who had departed OSTP before the final release but had led earlier drafting); and Ami Fields-Meyer, Chief of Staff for the Science and Society division.[^17][^18]
Following her departure from OSTP in February 2023, Nelson returned to the Institute for Advanced Study and continued to be a prominent public voice on AI governance. In 2023 TIME magazine named her to its inaugural list of the 100 most influential people in AI, citing her role in producing the Blueprint.[^19]
The Blueprint's substantive core consists of five principles, each presented with a short rights-style statement followed by a longer expository section and then a corresponding section in the technical companion. The five principles are listed in the order given in the document.[^1][^4]
The first principle states that "you should be protected from unsafe or ineffective systems."[^4] It calls for automated systems to be developed with consultation from diverse communities, stakeholders, and domain experts; for pre-deployment testing, risk identification, and risk mitigation; and for ongoing monitoring to demonstrate that systems are safe and effective based on their intended use.[^4] The Blueprint specifically references avoiding systems that have not been demonstrated to work as intended and protecting against "inappropriate or irrelevant" data use.[^4] It also urges independent evaluation and public reporting where feasible.[^4]
The second principle declares that "you should not face discrimination by algorithms and systems should be used and designed in an equitable way."[^4] The Blueprint defines algorithmic discrimination as occurring when automated systems contribute to unjustified different treatment or impacts disfavoring people on protected bases such as race, color, ethnicity, sex (including pregnancy, childbirth, and related conditions, gender identity, intersex status, and sexual orientation), religion, age, national origin, disability, veteran status, genetic information, or any other classification protected by law.[^4]
To prevent algorithmic discrimination, the principle calls on designers, developers, and deployers of automated systems to take "proactive and continuous measures," including equity assessments during design, the use of representative data, accessibility considerations for people with disabilities, pre-deployment and ongoing disparity testing and mitigation, organizational oversight, and independent evaluation and public reporting.[^4]
The third principle states that "you should be protected from abusive data practices via built-in protections and you should have agency over how data about you is used."[^4] It frames data privacy as a foundational right and calls for built-in protections by default, including limits on the collection of personal data to that which is strictly necessary for the relevant context, and meaningful consent mechanisms that are not framed in confusing legalese.[^4]
The principle treats certain "sensitive domains" with heightened protection, naming health, work, education, criminal justice, and finance, and data about youth.[^4] It also addresses continuous surveillance and monitoring, stating that surveillance should not be used to monitor employees' activities or restrict workers' organizing rights, and that surveillance technologies should be subject to "heightened oversight" including a determination of whether use is permissible at all in given contexts such as education or housing.[^4]
The fourth principle declares that "you should know that an automated system is being used and understand how and why it contributes to outcomes that impact you."[^4] Designers, developers, and deployers of automated systems are called upon to provide accessible plain-language documentation including clear descriptions of overall system functioning, the role of automation, notice that such systems are in use, and notice of when significant use, key functionality, or the system's outcomes have changed.[^4]
The principle further calls for explanations of outcomes to be technically valid, meaningful, useful to the people who interact with the system, and tailored to the level of risk based on the context.[^4] The Blueprint frames notice and explanation as foundational to all of the other principles, since they enable individuals to exercise rights such as opting out, contesting decisions, or seeking redress.[^4]
The fifth principle states that "you should be able to opt out, where appropriate, and have access to a person who can quickly consider and remedy problems you encounter."[^4] It calls for opportunities to opt out of automated systems in favor of a human alternative, where appropriate, based on factors such as the high stakes of decisions involving sensitive domains and the existence of "broad consensus" that human alternatives should be available.[^4]
The principle further calls for timely access to a fallback human reviewer and to an appropriate, accessible, and effective process for raising concerns, getting timely consideration, and obtaining remedy. In sensitive domains the Blueprint sets a particularly high bar, requiring "additional human oversight and safeguards" tailored to the relevant context such as criminal justice, employment, education, and health.[^4]
The Blueprint is paired with a separate companion document titled From Principles to Practice: A Technical Companion to the Blueprint for an AI Bill of Rights, released alongside the main document on October 4, 2022.[^8] OSTP described the companion as "a handbook for anyone seeking to incorporate these protections into policy and practice."[^4]
For each of the five principles, the companion is organized into three parts: a "Why This Principle Is Important" section describing real-world harms motivating the principle; a "What Should Be Expected of Automated Systems" section setting out detailed expectations from designers, developers, and deployers; and a "How These Principles Can Move Into Practice" section providing concrete real-life examples of laws, policies, and technical or organizational practices that already reflect the principle.[^8]
The examples in the companion are drawn from federal agencies, state and local governments, industry, civil society, and academia. They include items such as model AI risk-management practices, agency rules, internal corporate governance frameworks, and independent algorithmic audits. OSTP explicitly stated that the inclusion of an example was not an endorsement and that exclusion was not a criticism.[^8]
The Blueprint applies to "automated systems" rather than to AI as a narrow category. OSTP chose this broader term because, in its words, "the technical capabilities and specific definitions of such systems change with the speed of innovation."[^4] In the Blueprint's definition, an automated system is "any system, software, or process that uses computation as whole or part of a system to determine outcomes, make or aid decisions, inform policy implementation, collect data or observations, or otherwise interact with individuals and/or communities."[^4]
The Blueprint's protections are explicitly scoped to automated systems "that have the potential to meaningfully impact the American public's rights, opportunities, or access to critical resources or services."[^4] Examples of such impacts identified in the document include civil rights, civil liberties, and privacy; equal opportunities in education, housing, credit, employment, and other programs; and access to critical resources or services such as health care, financial services, safety, social services, non-deceptive information, and government benefits.[^4]
Examples of systems the Blueprint identifies as in-scope include facial recognition and other biometric technologies, tenant screening tools, employment selection algorithms, credit scoring models, healthcare risk-stratification tools, and various forms of educational and government-benefits decision support. The Blueprint generally excludes most industrial or operational applications of AI that do not meaningfully impact individuals' rights, opportunities, or access.[^4]
Reception of the Blueprint was politically and ideologically divided. Civil-society and civil-rights advocates were broadly supportive of the principles, while industry and free-market voices raised concerns about chilling effects, and a parallel set of critics from advocacy groups argued the document did not go far enough.[^10]
On the supportive side, Janet Haven of the Data and Society Research Institute described the Blueprint as positioning AI governance "as a framework for understanding AI governance broadly as a civil-rights issue."[^10] The Algorithmic Justice League, founded by Joy Buolamwini, called it "an encouraging step in the right direction" toward algorithmic justice.[^10] Academic commentators including Brookings Institution analysts characterized the framework as a meaningful conceptual contribution that elevated rights as a governance frame.[^20]
Critics from the academic and civil-society sides nevertheless noted the lack of enforcement. Russell Wald of the Stanford Institute for Human-Centered Artificial Intelligence said the Blueprint "lacks teeth or any details on enforcement," and called for executive actions, congressional hearings, and legislation.[^10] Albert Fox Cahn of the Surveillance Technology Oversight Project argued that for the most harmful systems "we don't need a blueprint, we need bans."[^10]
Industry response was mixed. Daniel Castro of the Center for Data Innovation criticized the Blueprint as "all wrong," arguing that existing laws already addressed many of the same harms and warning that an aggressive U.S. approach could undermine competitiveness against China.[^10] The U.S. Chamber of Commerce voiced what observers described as forceful pushback against the framework, expressing concern that federal agencies and state and local "copycats" would treat the non-binding Blueprint as a de facto baseline for regulation.[^10][^21]
Legal commentators highlighted both the document's ambition and its non-binding character. A Brookings analysis described "opportunities and blind spots" in the Blueprint, praising its civil-rights framing but flagging gaps around national security, large-language-model risks, and enforcement.[^22] A peer-reviewed article in Minds and Machines characterized the Blueprint as "in search of enaction, at risk of inaction," arguing that without binding implementation it risked remaining aspirational.[^23]
The Biden administration treated the Blueprint as a foundational policy document. When President Biden signed Executive Order 14110, Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence, on October 30, 2023, the order explicitly cited the Blueprint as part of the prior work upon which it built, alongside the National Institute of Standards and Technology's AI Risk Management Framework.[^11]
The relationship between the two documents was complementary. The Blueprint focused on individual rights vis-à-vis automated systems, while the 2023 ai executive order organized a broader federal agenda spanning safety testing for foundation models, content authentication, immigration of AI talent, agency rulemaking, and federal use of AI. Section 7 of the executive order, on advancing equity and civil rights through AI, drew most directly on the Blueprint's framing.[^11] Commentators characterized the combination of the Blueprint, the NIST AI Risk Management Framework, and Executive Order 14110 as the central pillars of the Biden administration's AI governance approach, blending rights-based and risk-based modes.[^24]
In the absence of comprehensive federal AI legislation, the Blueprint influenced a wave of state activity. The most prominent example is the Colorado Artificial Intelligence Act (CAIA), signed into law in May 2024, which became the first U.S. state law to comprehensively address algorithmic discrimination in "high-risk" AI systems.[^12] Observers noted that the CAIA's structure (with developer and deployer obligations, impact assessments, and protections against algorithmic discrimination in consequential decisions) tracked the framing of the Blueprint's algorithmic discrimination and data privacy principles.[^12]
The Blueprint was also cited by other state lawmakers considering AI legislation in California, New York, Illinois, Connecticut, and elsewhere, and was referenced in joint enforcement statements by federal civil-rights and consumer-protection agencies. In April 2023, the Federal Trade Commission, the Consumer Financial Protection Bureau, the Equal Employment Opportunity Commission, and the U.S. Department of Justice's Civil Rights Division issued a Joint Statement on Enforcement Efforts Against Discrimination and Bias in Automated Systems that reflected and reinforced the Blueprint's algorithmic discrimination principle.[^25]
Internationally, the Blueprint became one of the most frequently cited U.S. policy documents in cross-jurisdictional comparisons of AI governance approaches, often juxtaposed with the European Union's AI Act, China's interim measures for generative AI (china interim measures), South Korea's AI Basic Act (korea ai basic act), and the gpai code of practice in Europe. Within the U.S. governance ecosystem the Blueprint was frequently discussed in tandem with the nist aria program at the National Institute of Standards and Technology, which is oriented toward empirical AI evaluation.
Following the inauguration of Donald Trump on January 20, 2025, the Blueprint's status changed materially. Although the Blueprint was a non-binding white paper rather than an executive order, its companion instrument, Executive Order 14110, was among the Biden-era executive orders rescinded within hours of Trump's inauguration. The new administration's January 20, 2025 directive characterized Biden's AI executive order, alongside others, as reflecting "unpopular, inflationary, illegal, and radical practices."[^26]
The Blueprint itself was removed from the active whitehouse.gov OSTP website shortly after the change in administration. Copies of the page were preserved on the Biden White House archive at bidenwhitehouse.archives.gov and in third-party archives, including a copy maintained by the ACLU of Massachusetts dated January 2025.[^27][^28] As a non-binding white paper, the Blueprint could be administratively withdrawn through removal from official websites and through subsequent policy actions; no formal rescission was required and none was issued.[^26]
On January 23, 2025, Trump signed Executive Order Removing Barriers to American Leadership in Artificial Intelligence, which directed the heads of executive departments and agencies to identify and revise or rescind policies, regulations, and guidance "promulgated, issued, or adopted" pursuant to the rescinded Biden order. While the executive order did not name the Blueprint, agencies that had cited the Blueprint or its principles as a basis for guidance found themselves subject to the new directive.[^29]
On July 23, 2025, the Trump administration released Winning the Race: America's AI Action Plan, a 28-page document setting out approximately 90 federal policy actions across three pillars: accelerating innovation, building American AI infrastructure, and leading in international diplomacy and security.[^30] The ai action plan embraced a deregulation-focused approach, including reviews of existing federal AI regulations, restrictions on federal funding to states with "burdensome" AI regulations, and a focus on removing what the administration called "ideological bias" in AI systems used by the federal government.[^30]
The AI Action Plan does not formally repudiate the Blueprint, but the policy framing it offers is sharply different. Where the Blueprint centered civil rights and individual protections against algorithmic harms, the AI Action Plan centers competitiveness, infrastructure, and freedom from regulation that may slow AI deployment.[^31] Some commentators have noted that certain elements of the Blueprint's vocabulary (such as accountability, transparency, and human oversight) persist in modified form in subsequent federal documents, but most observers agree that the Blueprint no longer functions as a guiding federal framework.[^32]
As of 2026, the Blueprint remains a frequently cited reference in academic literature, state legislative debates, and international comparative law, and is preserved on the Biden White House archive. Its operational role within federal policy ended with the change of administration in January 2025, but its conceptual influence on rights-based AI governance arguments has continued in both U.S. state legislative work and international policy discussions.[^12][^25]