Pith. sign in

REVIEW 3 major objections 7 minor 1 cited by

Responsible Artificial Intelligence (RAI) in U.S. Federal Government : Principles, Policies, and Practices

T0 review · 3 major / 7 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper argues that U.S. federal AI policy can be mapped onto five Responsible AI pillars—fairness, reliability and robustness, transparency, accountability, and privacy and security—and that the Census Bureau is turning that mapping…

desk verdict A useful but uneven position paper from Census Bureau staff: the policy map is mostly right, the project descriptions are genuinely new, but factual slips and an overclaimed toolkit undermine its reliability as a reference. read the letter →

arxiv 2502.03470 v1 pith:BRUX5G4C submitted 2025-01-12 cs.CY

Atul Rawal , Katie Johnson , Curtis Mitchell , Michael Walton , Diamond Nwankwo This is my paper · ORCID
classification cs.CY
keywords responsibleAIfederalpolicygovernanceExecutiveOrder14110CensusBureauassessmenttoolkitmodelcardstrustworthy
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This position paper argues that the many U.S. federal AI policies, from executive orders to OMB guidance and voluntary frameworks, can be read as expressions of five Responsible AI pillars. It then presents Census Bureau projects as evidence that these high-level principles can be put into practice, with a model card generator, an AI registry, and an RAI assessment toolkit that directs users to specific tools for their AI system and data. The paper's most concrete assertion is that the toolkit, though still under development, provides a solution for evaluating domain-specific AI systems that use title-protected data. A sympathetic reader would care because this is a rare attempt to show federal RAI policy flowing from executive guidance down to working technical tools inside a statistical agency.

What carries the argument

The five-pillar RAI framework is the organizing device: fairness, reliability and robustness, transparency, accountability, and privacy and security. The operational mechanism is the RAI assessment toolkit, a web-based application under development at the Census Bureau that asks a team about its AI model and data, maps the answers to relevant portions of the executive orders, OMB M-24-10, and the RAI pillars, and returns applicable tools such as homomorphic encryption, secure multiparty computation, SHAP, and LIME. Supporting that mechanism are the model card generator, which produces standardized documentation, and the AI registry, which stores model records centrally for transparency and accountability.

What would settle it

Run the RAI assessment toolkit on a deliberately privacy-sensitive statistical AI system and check whether its output flags the required Title 13 and CIPSEA privacy protections alongside any fairness or transparency suggestions; if the toolkit omits those legal requirements or disagrees sharply with a human expert audit, the claim that it operationalizes RAI for protected data would be refuted.

Watch

Extended reading notes

Core claim

The paper's central claim is that U.S. federal Responsible AI policy is not a scattered collection of requirements but a coherent set of five pillars, and that the Census Bureau is one place where these pillars are becoming working practice. It reads EO 13859, EO 13960, EO 14110, OMB M-24-10, the AI Bill of Rights, the NIST AI Risk Management Framework, and the GAO accountability framework as jointly requiring fairness, reliability and robustness, transparency, accountability, and privacy and security. On the practice side, it describes the Census Bureau's model card generator for documenting a model's data, architecture, performance, and compliance, plus an AI registry that centralizes model information for governance. The strongest claim is that the RAI assessment toolkit offers a route from principles to concrete metrics and tool choices for statistical agencies working with Title 13 or CIPSEA-protected data, although the paper provides no evaluation results for the toolkit.

Load-bearing premise

The load-bearing premise is that an under-development software toolkit can translate high-level RAI principles and legal texts into accurate, complete, and reliable metrics and tool recommendations for any federal AI system, since the paper offers no validation data for that translation.

Editorial extensions

If this is right

  • If the five-pillar mapping is right, federal agencies can audit their AI systems against a single shared checklist rather than a tangle of separate executive orders and memos.
  • If the RAI assessment toolkit works as described, teams working with protected statistical data can move from a legal requirement to a concrete privacy or explainability tool without becoming RAI specialists.
  • The Census Bureau's model card generator and AI registry offer a replicable template for transparency and accountability that other federal agencies could follow.
  • Because most of these policies are executive orders, the paper concludes that codifying RAI requirements into law would make federal protections more durable across administrations.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A fair next test, beyond the paper, would be comparing the toolkit's recommendations against a panel of human expert auditors on a set of Census AI use cases; the paper gives no evaluation data for this comparison.
  • The five-pillar mapping may extend to agencies outside statistics, but the toolkit's question set and rule mappings would likely need tailoring to each agency's legal context and data types.
  • The Census Bureau examples show process adoption, not measured outcomes; without post-deployment monitoring data, the RAI practices described are not evidence of reduced bias or improved trust.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 7 minor

Summary. This position paper surveys Responsible AI (RAI) principles and the U.S. federal regulatory landscape for AI, covering Executive Orders 13859, 13960, and 14110, OMB M-24-10, the NIST AI RMF, and the GAO Accountability Framework. It then describes three Census Bureau initiatives: a model card generator, an AI registry, and a RAI assessment toolkit that is described as currently under development. The paper's central claim is that this toolkit will operationalize RAI principles for federal statistical agencies working with Title-protected and CIPSEA-protected data, translating high-level policy into concrete metrics and tool recommendations.

Significance. If the RAI assessment toolkit delivered what Section 5 claims, it would be a practical contribution to RAI governance in federal statistical agencies, providing a bridge from principles to technical practice. The paper also usefully compiles recent federal AI policy documents and describes concrete Census Bureau projects, which is a relatively underexplored area in the RAI literature. However, the paper's regulatory overview contains several factual inaccuracies, and its strongest capability claim is about an unvalidated, in-development tool. The paper is therefore best read as a preliminary position paper rather than a settled account of federal RAI practice. Its strengths are its coverage of current policy documents and its explicit, concrete use cases from a major statistical agency.

major comments (3)
  1. [Section 1] The paragraph beginning 'Laws such as the U.S Algorithmic Accountability Act of 2019' contains three factual errors that undermine the reliability of the regulatory overview: EOs are cited as 'EO13960 and EO14410', but the correct number is EO 14110; the U.S. Algorithmic Accountability Act of 2019 was a proposed bill and is not enacted law, so it should not be described as a law that 'dictates' assessments; and the GDPR's 'Right to Explanation' is a contested interpretation of the regulation, not an unambiguous statutory consumer right. These should be corrected and appropriately hedged.
  2. [Section 4.2 and Section 5] The central capability claim is unsupported. Section 4.2 states the RAI assessment toolkit is 'currently under-development' and describes intended behavior in the future tense ('will focus', 'will direct users', 'would then suggest'), but Section 5 asserts that the toolkit 'provides a solution for the evaluation and assessment of domain specific AI systems utilizing title protected datasets.' The paper gives no information about the toolkit's rule mappings from policy text to RAI principles, its question set, its scoring or decision procedure, or any validation or evaluation data demonstrating that its recommendations are accurate, complete, or robust. Without such evidence, the paper cannot support a claim that the toolkit already operationalizes RAI; it can only claim that such a toolkit is being designed. This discrepancy must be resolved, either by removing the Section 5 claim or by adding concrete design and evaluation details.
  3. [Section 3.1] The discussion of EO 13859 states that the order 'substantially increase[d] funding' for AI research, including specific dollar amounts for NSF, DoE, NIH, and agriculture. This conflates an executive order's direction to prioritize existing investments with actual appropriations, which are made by Congress. Please clarify that EO 13859 directed agencies to prioritize AI research and that the cited funding levels were part of broader appropriations or agency plans, not direct appropriations by the EO.
minor comments (7)
  1. [Section 1] The phrase 'European Unions, General Data Protection Regulation' contains a grammatical error; it should be 'European Union's General Data Protection Regulation'.
  2. [Section 2] The definition of explainability in the paragraph on Transparency quotes Rawal et al. but contains a typo: 'the system to capable of allowing' should be 'the system is capable of allowing'.
  3. [Section 2] In the paragraph on Privacy and security, 'Title and CIPSEA' is used without explaining that 'Title' refers to Title 13 of the U.S. Code and CIPSEA refers to the Confidential Information Protection and Statistical Efficiency Act; this should be spelled out at first use.
  4. [Section 3.1] The text refers to 'the AI Bill of rights' and later 'The AI Bill of rights, released in October 2022'; the official name is the 'Blueprint for an AI Bill of Rights', and it should be cited consistently with that title.
  5. [Section 4.1] The sentence 'Teams/divisions using AI/ML products within their research or duties are be able to submit their models' contains a grammar error: 'are be able' should be 'are able'.
  6. [Section 1] The paper cites reference [7] (MacCarthy, 'An examination of the algorithmic accountability act of 2019') to support the claim about the Act's content; it would be more appropriate to cite the bill itself or a more direct source, especially since the Act is only proposed legislation.
  7. [Section 2] The phrase 'with almost little to no human involvement' in the Introduction is awkward; consider 'with little to no human involvement'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: this is a position paper with no derivation chain, no fitted parameter presented as a prediction, and only one expository self-citation that is not load-bearing.

full rationale

The paper makes no formal derivation or empirical prediction, so there is no reduction of outputs to inputs. Its central claims are: five RAI pillars organize federal policy; Executive Orders, the OMB M-24-10 memo, NIST AI RMF, and GAO guidance contain corresponding requirements; and an under-development Census RAI assessment toolkit will help operationalize these principles. The Section 5 mapping from EOs to pillars is a qualitative interpretation of policy text, not a computed result, and the toolkit claim in Sections 4.2 and 5 is an unsupported capability claim rather than a circular one: the toolkit is described as 'currently under-development' and no evidence is offered for the accuracy of its rule mappings, question set, or tool recommendations. The only self-citation is the Rawal et al. definition of explainability in Section 2, used for exposition; the paper's argument does not derive any load-bearing conclusion from that definition, and no uniqueness theorem, ansatz, or fitted value is imported from prior work. Factual slips such as the incorrect EO number and treating a proposed bill as law are correctness risks, not circularity. Accordingly, the derivation chain is self-contained in the sense that it has no circular step.

Assumptions & free parameters 0 free parameters · 2 assumptions · 0 invented entities

The paper advances no quantitative or empirical central claim, so it introduces no fitted parameters and no invented scientific entities. It does rely on two expository assumptions: the five-pillar taxonomy of RAI is valid and complete, and official federal documents are accurately summarized and authoritative. Neither is defended, but neither is load-bearing in the sense of a derivational result.

assumptions (2)
  • domain assumption The five RAI pillars (fairness, reliability and robustness, transparency, accountability, privacy and security) form a valid and complete decomposition of responsible AI.
    Invoked throughout Section 2 and used in Section 5 to map every executive order and memo onto one or more pillars; the paper gives no independent justification for choosing exactly these five.
  • domain assumption Official federal documents (executive orders, OMB memos, NIST and GAO frameworks) are accurately summarized and authoritative for describing current RAI requirements.
    Section 3 treats these documents as ground truth for the policy landscape; the paper does not verify against implementation outcomes beyond citing a 2023 GAO report.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Responsible Artificial Intelligence (RAI) in U.S. Federal Government : Principles, Policies, and Practices." pith.science (2026). https://pith.science/paper/BRUX5G4C

@misc{pith2026250203470,
  author       = {Pith},
  title        = {Pith review of: Responsible Artificial Intelligence (RAI) in U.S. Federal Government : Principles, Policies, and Practices},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BRUX5G4C}},
  note         = {Machine review of arXiv:2502.03470}
}
read the original abstract

Artificial intelligence (AI) and machine learning (ML) have made tremendous advancements in the past decades. From simple recommendation systems to more complex tumor identification systems, AI/ML systems have been utilized in a plethora of applications. This rapid growth of AI/ML and its proliferation in numerous private and public sector applications, while successful, has also opened new challenges and obstacles for regulators. With almost little to no human involvement required for some of the new decision-making AI/ML systems, there is now a pressing need to ensure the responsible use of these systems. Particularly in federal government use-cases, the use of AI technologies must be carefully governed by appropriate transparency and accountability mechanisms. This has given rise to new interdisciplinary fields of AI research such as \textit{Responsible AI (RAI)}. In this position paper we provide a brief overview of development in RAI and discuss some of the motivating principles commonly explored in the field. An overview of the current regulatory landscape relating to AI is also discussed with analysis of different Executive Orders, policies and frameworks. We then present examples of how federal agencies are aiming for the responsible use of AI, specifically we present use-case examples of different projects and research from the Census Bureau on implementing the responsible use of AI. We also provide a brief overview for a Responsible AI Assessment Toolkit currently under-development aimed at helping federal agencies operationalize RAI principles. Finally, a robust discussion on how different policies/regulations map to RAI principles, along with challenges and opportunities for regulation/governance of responsible AI within the federal government is presented.

Figures

Figures reproduced from arXiv: 2502.03470 by the authors.

Figure 1
Figure 1. (A) Yearly publications for responsible, ethical and trustworthy AI. (Data derived from [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. A timeline of AI related guidance/policies from the U.S government. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. An overview of the RAI assessment toolkit framework. [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Toward Effective AI Governance: A Review of Principles

    cs.SE 2025-05 reject novelty 4.0 of 10

    A rapid tertiary review of nine AI governance reviews finds a focus on high-level frameworks and principles, with little concrete guidance on governance mechanisms.

Reference graph

Works this paper leans on

28 extracted references · 23 canonical work pages · cited by 1 Pith paper

  1. [1]

    Explainable artificial intelligence (xai): Concepts, taxonomies, opportunities and challenges toward responsible ai,

    A. B. Arrieta, N. Díaz-Rodríguez, J. Del Ser, A. Bennetot, S. Tabik, A. Barbado, S. García, S. Gil-López, D. Molina, R. Benjaminset al., “Explainable artificial intelligence (xai): Concepts, taxonomies, opportunities and challenges toward responsible ai,” Information Fusion, vol. 58, pp. 82–115, 2020

    work page 2020

  2. [2]

    European union regulations on algorithmic decision-making and a “right to explanation

    B. Goodman and S. Flaxman, “European union regulations on algorithmic decision-making and a “right to explanation”,” AI magazine, vol. 38, no. 3, pp. 50–57, 2017

    work page 2017

  3. [3]

    Amazon scraps secret ai recruiting tool that showed bias against women,

    J. Dastin, J. Weber, and M. Dickerson, “Amazon scraps secret ai recruiting tool that showed bias against women,” Reuters, 2018

    work page 2018

  4. [4]

    Facebook’s ad-serving algorithm discriminates by gender and race,

    K. Hao, “Facebook’s ad-serving algorithm discriminates by gender and race,” MIT Technology Review, 2019. [Online]. Available: https://www.technologyreview.com/2019/04/05/1175/ facebook-algorithm-discriminates-ai-bias/

    work page 2019

  5. [5]

    Millions of black people affected by racial bias in health-care algorithms,

    H. Ledford, “Millions of black people affected by racial bias in health-care algorithms,” Nature, vol. 574, no. 7780, pp. 608–609, 2019. [Online]. Available: https://www.ncbi.nlm.nih.gov/ pubmed/31664201

    work page arXiv 2019

  6. [6]

    The eu general data protection regulation (gdpr),

    P. V oigt and A. V on dem Bussche, “The eu general data protection regulation (gdpr),”A Practical Guide, 1st Ed., Cham: Springer International Publishing, vol. 10, p. 3152676, 2017

    work page 2017

  7. [7]

    An examination of the algorithmic accountability act of 2019,

    M. MacCarthy, “An examination of the algorithmic accountability act of 2019,”SSRN Electronic Journal, 2019

    work page 2019

  8. [8]

    Responsible AI by Design in Practice

    R. Benjamins, A. Barbado, and D. Sierra, “Responsible ai by design in practice,” arXiv preprint arXiv:1909.12838, 2019

    work page Pith review arXiv 1909

Show all 28 references
  1. [9]

    Ethics and responsible ai deploy- ment,

    P. Radanliev, O. Santos, A. Brandon-Jones, and A. Joinson, “Ethics and responsible ai deploy- ment,” Frontiers in Artificial Intelligence, vol. 7, p. 1377011, 2024

    2024

  2. [10]

    Principles to practices for responsible ai: closing the gap,

    D. Schiff, B. Rakova, A. Ayesh, A. Fanti, and M. Lennon, “Principles to practices for responsible ai: closing the gap,” arXiv preprint arXiv:2006.04707, 2020

    2006 arXiv

  3. [11]

    Responsible ai pattern catalogue: A multivocal literature review,

    Q. Lu, L. Zhu, X. Xu, J. Whittle, D. Zowghi, and A. Jacquet, “Responsible ai pattern catalogue: A multivocal literature review,”arXiv preprint arXiv:2209.04963, 2022

    2022 arXiv

  4. [12]

    Responsible ai pattern catalogue: A collection of best practices for ai governance and engineering,

    ——, “Responsible ai pattern catalogue: A collection of best practices for ai governance and engineering,” ACM Computing Surveys, vol. 56, no. 7, pp. 1–35, 2024

    2024

  5. [13]

    Socially responsible ai algorithms: Issues, purposes, and challenges,

    L. Cheng, K. R. Varshney, and H. Liu, “Socially responsible ai algorithms: Issues, purposes, and challenges,” Journal of Artificial Intelligence Research, vol. 71, pp. 1137–1181, 2021

    2021

  6. [14]

    A rapid review of responsible ai frame- works: How to guide the development of ethical ai,

    V . S. Barletta, D. Caivano, D. Gigante, and A. Ragone, “A rapid review of responsible ai frame- works: How to guide the development of ethical ai,” in Proceedings of the 27th International Conference on Evaluation and Assessment in Software Engineering, 2023, pp. 358–367

    2023

  7. [15]

    A framework for fairness: A systematic review of existing fair ai solutions,

    B. Richardson and J. E. Gilbert, “A framework for fairness: A systematic review of existing fair ai solutions,” arXiv preprint arXiv:2112.05700, 2021

    2021 arXiv

  8. [16]

    Fair ai: Challenges and opportunities,

    S. Feuerriegel, M. Dolata, and G. Schwabe, “Fair ai: Challenges and opportunities,” Business & information systems engineering, vol. 62, pp. 379–384, 2020

    2020

  9. [17]

    Towards guaranteed safe ai: A framework for ensuring robust and reliable ai systems,

    D. Dalrymple, J. Skalse, Y . Bengio, S. Russell, M. Tegmark, S. Seshia, S. Omohundro, C. Szegedy, B. Goldhaber, N. Ammann et al., “Towards guaranteed safe ai: A framework for ensuring robust and reliable ai systems,” arXiv preprint arXiv:2405.06624, 2024

    2024 arXiv

  10. [18]

    In ai we trust: ethics, artificial intelligence, and reliability,

    M. Ryan, “In ai we trust: ethics, artificial intelligence, and reliability,” Science and Engineering Ethics, vol. 26, no. 5, pp. 2749–2767, 2020

    2020

  11. [19]

    How to make ai more reliable,

    O. Kosheleva and V . Kreinovich, “How to make ai more reliable,”UTEP Departmental Technical Report, 2024

    2024

  12. [20]

    Recent advances in trustworthy explainable artificial intelligence: Status, challenges, and perspectives,

    A. Rawal, J. McCoy, D. B. Rawat, B. M. Sadler, and R. S. Amant, “Recent advances in trustworthy explainable artificial intelligence: Status, challenges, and perspectives,” IEEE Transactions on Artificial Intelligence, vol. 3, no. 6, pp. 852–866, 2021

    2021

  13. [21]

    Responsible artificial intelligence: A structured literature review,

    S. Goellner, M. Tropmann-Frick, and B. Brumen, “Responsible artificial intelligence: A structured literature review,”arXiv preprint arXiv:2403.06910, 2024. 10

    2024 arXiv

  14. [22]

    Maintaining american leadership in artificial intelligence,

    “Maintaining american leadership in artificial intelligence,” Executive Order 13859, 2019. [Online]. Available: https://www.federalregister.gov/d/2019-02544

    2019

  15. [23]

    Promoting the use of trustworthy artificial intelligence in the federal government,

    “Promoting the use of trustworthy artificial intelligence in the federal government,” Executive Order 13960, 2020. [Online]. Available: https://www.federalregister.gov/d/2020-27065

    2020

  16. [24]

    Safe, secure, and trustworthy development and use of artificial intelligence,

    “Safe, secure, and trustworthy development and use of artificial intelligence,” Executive Order 14110, 2023. [Online]. Available: https://www.federalregister.gov/d/2023-24283

    2023

  17. [25]

    Advancing governance, innovation, and risk management for agency use of artificial intelligence,

    “Advancing governance, innovation, and risk management for agency use of artificial intelligence,” OMB M-24-10, 2024. [On- line]. Available: https://www.whitehouse.gov/wp-content/uploads/2024/03/ M-24-10-Advancing-Governance-Innovation-and-Risk-Management-for-Agency-Use-of-Art...

    2024

  18. [26]

    Blueprint for an ai bill of rights,

    “Blueprint for an ai bill of rights,” 2022. [Online]. Available: https://www.whitehouse.gov/ostp/ ai-bill-of-rights/about-this-document/

    2022

  19. [27]

    Artificial intelligence risk management framework (ai rmf 1.0),

    E. Tabassi, “Artificial intelligence risk management framework (ai rmf 1.0),” 2023-01-26 05:01:00 2023. [Online]. Available: https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id= 936225

    2023

  20. [28]

    Artificial intelligence:an accountability framework for federal agencies and other entities,

    “Artificial intelligence:an accountability framework for federal agencies and other entities,” GAO-21-519SP, 2021. [Online]. Available: https://www.gao.gov/products/gao-21-519sp 11

    2021

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.