CoChat
Sign Up
Cited by 1Ethics and Social Impacts of AILegal Issues in EducationArtificial Intelligence in Education

The Due Process Deficit: Auditing AI Governance in U.S. Higher Education

Open access
Teleki, Maria, Choi, Anna Seo Gyeong, Duray, Anne, Liu, Haoran, Zhang, Junyan, Dong, Xiangjue, Da Silva, Dilma, Koenecke, Allison +1 more·2026·Proceedings of the 2026 ACM Conference on Fairness, Accountability, and Transparency
DOI 10.1145/3805689.3812328
Open this paper in CoChat — free account

Abstract

U.S. universities have met Artificial Intelligence (AI) with policy - revised honor codes, AI use guidelines, and detection tools to enforce them - but the procedural infrastructure that would let affected parties understand, contest, and seek redress for AI-related decisions has not kept pace. We present ACAI-US79, a benchmark dataset allowing for institutional auditing of AI governance across 79 U.S. universities spanning R1, R2, and teaching-focused institutions, and ACAI, the Academic AI Capacity Index, an interpretable measure of the public legibility of governance structures that allocate authority and accountability. Unlike prior audits focused on pedagogical guidance at elite or Rl-only samples, ACAI shifts the analytic frame to procedural accountability, evaluating four domains — policy clarity, faculty support, feedback mechanisms, and AI detection tool governance — through time-bounded review of institutionally authoritative materials. We find that governance capacity is systematically concentrated in rule articulation while mechanisms for participation, feedback, and procedural constraint are underdeveloped, and that AI research intensity is weakly correlated with AI governance capacity - i.e., we find that technical leadership does not translate into procedural accountability. We further show that LLM-based audits produce unstable institutional rankings, demonstrating the continued necessity of human judgment in high-stakes audit contexts. We release the dataset, audit instrument, and public website at http://acai-us79.org/ to support transparency, critique, and institutional self-reflection, laying empirical groundwork for accountable AI governance in higher education.

AI snapshot

Written from the paper’s full text.

TL;DR

The paper audits publicly articulated AI governance across 79 U.S. universities and introduces ACAI, an interpretable index of procedural accountability across policy clarity, faculty support, feedback mechanisms, and detection-tool governance. It finds governance capacity concentrated in rule articulation rather than participation or due process, weakly correlated with AI research output, and not reliably reproducible by LLM-based audits.

Key findings

  • Governance capacity is systematically skewed toward rule articulation: across institutional types, sizes, research intensities, and regions, Policy Clarity (A) and Faculty Support (B) outscore Feedback Mechanisms (C) and Detection Tools (D), leaving participation and due process undersupported.
  • AI governance capacity is weakly correlated with AI research output: across the N=35 institutions ranked by both ACAI and CSRankings AI, Pearson r=0.17, Spearman ρ=0.19 (95% bootstrap CI [-0.15, 0.47]), Kendall τ=0.12. The highest ACAI score is University of New Hampshire at 81.82, followed by Portland State University and Stanford University at 80.30.
  • LLM-based audits only partially reproduce human governance judgments: across temperatures, LLM rankings show moderate to strong ordinal agreement with human rankings (ρ=0.54–0.62), but mean absolute rank errors are 15–17 positions out of 79, with top-quartile misrankings up to 64–72 positions and bottom-quartile discrepancies of 0–8 positions.
  • ACAI rankings are robust to weighting choices and individual annotators: across four weighting schemes, Spearman ρ=0.93–0.99 and Pearson r=0.93–0.99; leave-one-annotator-out recomputation remains strongly correlated with the full index (ρ=0.77–0.87, r=0.83–0.87). Inter-annotator agreement is modest at item level (Krippendorff’s α=0.24) and improves at domain level (α=0.30).

Methods

  • The authors construct ACAI-US79, a benchmark dataset of 79 U.S. universities spanning R1, R2, and teaching/liberal arts institutions, selected purposively across U.S. Census regions and Carnegie categories. For each university, they collect institutionally authoritative links across seven categories and annotate governance indicators under time-bounded review.
  • Human annotation was conducted by paid Prolific annotators, with each university independently annotated by three annotators; two computational PhD students independently annotated a subset of 10 universities for expert validation. ACAI aggregates four domains—Policy Clarity, Faculty Support, Feedback Mechanisms, Detection Tools—using scores of 1.0, 0.5, or 0.0, weighted baseline wA=2, wB=4, wC=3, wD=2, and scaled to [0,100].
  • The authors compare ACAI against CSRankings AI research output and conduct an LLM study with gpt-5.2-2025-12-11, running three independent runs at temperatures τ ∈ {0.5, 1.0, 1.5} and requiring validated JSON outputs.

Limitations

  • The paper states that the sample is limited to 79 universities and is USA-centric, so it may not generalize globally; it relies on publicly available institutional data, omitting informal practices or internal decision-making; and it excludes formal but non-public materials such as intranet pages, learning management systems, or password-protected policy portals, meaning ACAI should be interpreted as a lower bound on governance capacity.
  • The text makes evident that subgroup comparisons are not confirmatory: the audit sample is purposive rather than probabilistic, subgroup categories are administratively defined and analytically coarse, and multiple overlapping contrasts are examined without correction.
  • The text also makes evident that annotation agreement is modest and variable, and that LLM-generated rankings are unstable and frequently severely misrank individual institutions, limiting automated audit reliability.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1145/3805689.3812328

Cites

41 of 42 references verified0 citation errors · 1 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

49 of 87 cited sources checked · 27 supported · 0 flagged · accuracy not yet measured

  • Premisecitation
    § 2.1 AI in Higher Education · p. 4

    Detection systems disproportionately misclassify the work of non-native English speakers and neurodivergent writers.

    “False positive rates vary widely across tools, and detection systems disproportionately misclassify the work of non-native English speakers and neurodivergent writers [43, 77].”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [43] [43] Weixin Liang, Mert Yuksekgonul, Yining Mao, Eric Wu, and James Zou. 2023. GPT detectors are biased against non-native English writers. Patterns 4, 7 (2023), 100779. doi:10.1016/j.patter.2023.100779 - source

      The section “GPT detectors exhibit bias against non-native English authors” reports that seven detectors mislabeled more than half of the non-native writers’ TOEFL essays as AI-generated, while accurately classifying the comparison essays. The paper does not address neurodivergent writers, so it supports only the non-native-English portion of the attributed claim. (LLM, conf=0.88)

      Experimental — accuracy not yet measured
    • cites [77] [77] Debora Weber-Wulff, Alla Anohina-Naumeca, Sonja Bjelobaba, Tomáš Folt` ynek, Jean Guerrero-Dib, Olumide Popoola, Petr Šigut, and Lorna Waddington. 2023. Testing of detection tools for AI-generated text. International Journal for Educational Integrity 19, 1 (2023), 1-39. - source

      The paper reports that false-positive rates varied across tools (Human-Generated Content, p. 26) and that machine-translated texts had more false positives, putting L2 students who translate their own work at risk (False accusations, p. 20; Human-Generated Content, p. 26). This supports the concern for that specific group and condition, but not a general claim about all non-native English speakers. Our copy does not address neurodivergent writers. (LLM, conf=0.90)

      Experimental — accuracy not yet measured
  • Premisecitation
    § 2.1 AI in Higher Education · p. 4

    False positive rates vary widely across AI detection tools.

    “False positive rates vary widely across tools, and detection systems disproportionately misclassify the work of non-native English speakers and neurodivergent writers [43, 77].”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [43] [43] Weixin Liang, Mert Yuksekgonul, Yining Mao, Eric Wu, and James Zou. 2023. GPT detectors are biased against non-native English writers. Patterns 4, 7 (2023), 100779. doi:10.1016/j.patter.2023.100779 - source

      In “GPT detectors exhibit bias against non-native English authors” (page unavailable), the paper reports an average false-positive rate of 61.3% across seven detectors and shows that they did not flag identical sets of TOEFL essays. The Figure 1 caption also describes standard-deviation error bars across detectors. These passages support variation, but they do not establish how widely the individual tools’ false-positive rates differed. (LLM, conf=0.78)

      Experimental — accuracy not yet measured
    • cites [77] [77] Debora Weber-Wulff, Alla Anohina-Naumeca, Sonja Bjelobaba, Tomáš Folt` ynek, Jean Guerrero-Dib, Olumide Popoola, Petr Šigut, and Lorna Waddington. 2023. Testing of detection tools for AI-generated text. International Journal for Educational Integrity 19, 1 (2023), 1-39. - source
  • Premisecitation
    § 2.1 AI in Higher Education · p. 4

    Institutional guidance on how AI detection outputs should be interpreted, weighted, or contested remains largely absent.

    “Yet institutional guidance on how detection outputs should be interpreted, weighted, or contested remains largely absent [47, 76].”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [47] [47] Nora McDonald, Aditya Johri, Areej Ali, and Aayushi Hingle Collier. 2025. Generative artificial intelligence in higher education: Evidence from an analysis of institutional policies and guidelines. Computers in Human Behavior: Artificial Humans 3 (2025), 100121. - source

      The Overview (p. 8) reports that 23% of the 116 institutions provided guidance on using GenAI detection tools, consistent with such guidance being uncommon. Section 4.2.2 (p. 12) describes cautions about detector reliability, and §4.3.1 (p. 14) gives one institution’s rule against pursuing cases based solely on detector results. The paper does not establish how often institutions guide faculty on interpreting or weighing specific outputs, or on contesting them. (LLM, conf=0.78)

      Experimental — accuracy not yet measured
    • cites [76] [76] Hui Wang, Anh Dang, Zihao Wu, and Son Mac. 2024. Generative AI in higher education: Seeing ChatGPT through universities' policies, resources, and guidelines. Computers and Education: Artificial Intelligence 7 (2024), 100326. - source

      Section 5.2.3 (p. 8) reports that none of the universities studied regarded AI detectors as completely reliable or supported using them to determine plagiarism. That supports caution about the weight given to detection outputs, but the paper does not establish that guidance on interpreting or contesting specific outputs is largely absent; it also documents some institutional guidance on detector use. (LLM, conf=0.82)

      Experimental — accuracy not yet measured
  • Premisecitation
    § 2.1 AI in Higher Education · p. 4

    AI has achieved rapid, widespread adoption in higher education.

    “AI has achieved rapid, widespread adoption in higher education. Within two years of ChatGPT's release, over 80% of students were using AI academically [21], and a 2025 survey found that 85% had used AI for coursework in the past year [28].”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [21] [21] Zara Contractor and Germán Reyes. 2025. Generative AI in Higher Education: Evidence from an Elite College. arXiv:2508.00717 [econ.GN] https://arxiv.org/abs/2508.00717 - source
    • cites [28] [28] Colleen Flaherty. 2025. How AI Is Changing-Not 'Killing'-College. Inside Higher Ed (29 Aug. 2025). https://www.insidehighered.com/ news/students/academics/2025/08/29/survey-college-students-views-ai -

      Not checked in the free run.

  • Premisecitation
    § 3.2 ACAI: The Academic AI Capacity Index · p. 9

    Institutional guidance on AI detection tools is often limited, ambiguous, or silent on procedural constraints.

    “Despite these risks, institutional guidance on detection tools is often limited, ambiguous, or silent on procedural constraints [47]”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [47] [47] Nora McDonald, Aditya Johri, Areej Ali, and Aayushi Hingle Collier. 2025. Generative artificial intelligence in higher education: Evidence from an analysis of institutional policies and guidelines. Computers in Human Behavior: Artificial Humans 3 (2025), 100121. - source

      The Overview (p. 8) reports that only 27 institutions (23%) provided guidance on using GenAI detection tools, supporting the claim that such guidance is often limited. Section 4.2.2 (p. 12) describes cautions about reliability, but does not assess whether guidance is ambiguous or silent on procedural constraints. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
  • May combine two claims. 58
  • Classified as probably not a substantial claim. 51
  • The quote only partly states this claim. 2

  • May combine two claims.

    Premisecitation
    § 1 Introduction · p. 3

    Prior empirical audits of university AI governance have focused on policy presence among elite or research-intensive samples, examining primarily the pedagogical dimension.

    “Prior empirical audits of university AI governance have focused on policy presence among elite or researchintensive samples [47, 50, 76, 80] (Table 2, Appendix B), examining primarily the pedagogical dimension: how universities communicate expectations about AI use in coursework.”
    • Also stated in: § 2.2 Auditing Institutional AI Governance: Ethics and Accountability · p. 4 · plain
      “Previous studies analyze AI governance across only elite universities (Table 2, Appendix B) - 'Big 10' [80], 'Top 100' [76], 'Top 50' [50], or the 116 R1 universities in the U.S. [47] - focusing only on the pedagogical dimension: how universities communicate expectations about AI use in coursework.”
    Quote in the paperSubstantial claim3 of 4 sources checked
    • cites [47] [47] Nora McDonald, Aditya Johri, Areej Ali, and Aayushi Hingle Collier. 2025. Generative artificial intelligence in higher education: Evidence from an analysis of institutional policies and guidelines. Computers in Human Behavior: Artificial Humans 3 (2025), 100121. - source
    • cites [50] [50] Benjamin Luke Moorhouse, Marie Alina Yeo, and Yuwei Wan. 2023. Generative AI tools and assessment: Guidelines of the world's top-ranking universities. Computers and Education Open 5 (2023), 100151. - source
    • cites [76] [76] Hui Wang, Anh Dang, Zihao Wu, and Son Mac. 2024. Generative AI in higher education: Seeing ChatGPT through universities' policies, resources, and guidelines. Computers and Education: Artificial Intelligence 7 (2024), 100326. - source
    • cites [80] [80] Chuhao Wu, He Zhang, and John M. Carroll. 2024. AI Governance in Higher Education: Case Studies of Guidance at Big Ten Universities. arXiv:2409.02017 [cs.HC] https://arxiv.org/abs/2409.02017 - source

      The Methods and §3.1 describe a study of AI guidelines at large, research-intensive Big Ten universities, supporting the sample characterization. But the Introduction explicitly says this study extends beyond teaching and learning to examine AI governance at the organizational and community levels; the description of earlier work as focused on guideline publication and pedagogical topics does not fully describe this cited study. The indexed passages have section labels but no page numbers. (LLM, conf=0.88)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Premisecitation
    § 2.3 Legal Grounding · p. 5

    Meaningful accountability requires information, explanation, and the possibility of consequences.

    “Meaningful accountability requires information, explanation, and the possibility of consequences [11, 12]”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites [11] [11] Marcus Alphons Petrus Bovens, Robert E Goodin, and Thomas Schillemans. 2014. The Oxford handbook of public accountability . Oxford handbooks. -

      Not checked in the free run.

    • cites [12] [12] Madalina Busuioc. 2021. Accountable artificial intelligence: Holding algorithms to account. Public administration review 81, 5 (2021), 825-836. - source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 2.1 AI in Higher Education · p. 4

    65% of STEM academic scientists report using generative AI in teaching or research activities.

    “Adoption extends to instructors as well: 65% of STEM academic scientists report using generative AI in teaching or research activities [5].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [5] [5] Wenceslao Arroyo-Machado, Jinghuan Ma, Tipeng Chen, Timothy P Johnson, Shaika Islam, Lesley Michalegko, and Eric Welch. 2025. Generative AI and academic scientists in US universities: Perception, experience, and adoption intentions. PloS one 20, 8 (2025), e0330416. - source

      The Abstract and Discussion say that 65% of surveyed STEM academic scientists used generative AI in teaching or research. However, the Results distinguish 65% who used it for any purpose from 40% who used it in teaching or research; the remaining 25% used it for other functions. The cited paper is internally inconsistent on the figure in the claim. (LLM, conf=0.86)

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation
    § 3.2 ACAI: The Academic AI Capacity Index · p. 9

    Prior research has documented significant technical limitations and bias in AI detection systems, as well as their potential to produce false positives with serious consequences for students.

    “Prior research - shown in Table 4 - has documented significant technical limitations and bias in AI detection systems, as well as their potential to produce false positives with serious consequences for students [43, 76].”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [43] [43] Weixin Liang, Mert Yuksekgonul, Yining Mao, Eric Wu, and James Zou. 2023. GPT detectors are biased against non-native English writers. Patterns 4, 7 (2023), 100779. doi:10.1016/j.patter.2023.100779 - source
    • cites [76] [76] Hui Wang, Anh Dang, Zihao Wu, and Son Mac. 2024. Generative AI in higher education: Seeing ChatGPT through universities' policies, resources, and guidelines. Computers and Education: Artificial Intelligence 7 (2024), 100326. - source

      Section 5.2.3 (p. 8) reports that universities do not regard AI detectors as completely reliable and describes how minor paraphrasing can substantially reduce detection accuracy. Section 6.1 (p. 11) urges caution when accusing students. The cited paper does not document bias in detection systems or specifically link false positives to serious consequences for students. (LLM, conf=0.86)

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim.

    Premisecitation
    § 3.2 ACAI: The Academic AI Capacity Index · p. 9

    Scholarship on algorithmic governance emphasizes that accountability requires institutionalized mechanisms for participation, feedback, and revision over time.

    “More broadly, scholarship on algorithmic governance emphasizes that accountability requires institutionalized mechanisms for participation, feedback, and revision over time.”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [49] [49] Jacob Metcalf, Emanuel Moss, Elizabeth Anne Watkins, Ranjit Singh, and Madeleine Clare Elish. 2021. Algorithmic impact assessments and accountability: The co-construction of impacts. In Proceedings of the 2021 ACM conference on fairness, accountability, and transparency . 735-746. FAccT '26, June 25-28, 2026, Montreal, QC, Canada - source
    • cites [65] [65] Andrew D Selbst. 2021. An institutional view of algorithmic impact assessments. Harv. JL & Tech. 35 (2021), 117. -

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 3.1 ACAI-US79: A Benchmarking Dataset · p. 7

    The seven link categories correspond to distinct organizational surfaces through which governance is commonly articulated.

    “These categories correspond to distinct organizational surfaces through which governance is commonly articulated [80], initialized by a manual author review of a small set of university policies, and then iterated on during our search process which is detailed in Appendix F.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [80] [80] Chuhao Wu, He Zhang, and John M. Carroll. 2024. AI Governance in Higher Education: Case Studies of Guidance at Big Ten Universities. arXiv:2409.02017 [cs.HC] https://arxiv.org/abs/2409.02017 - source

      Section 4.1, “Multi-Unit Governance of AI” (page unavailable), reports that different university units publish AI guidance and have distinct roles in governance. That supports the idea of different organizational surfaces, but our copy does not establish that the citing paper’s seven link categories correspond to those units. (LLM, conf=0.84)

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation
    § 3.2 ACAI: The Academic AI Capacity Index · p. 9

    Prior work has shown heterogeneity in how universities define and communicate AI-related expectations, with many institutions relying on vague or decentralized guidance.

    “Prior work has shown heterogeneity in how universities define and communicate AI-related expectations, with many institutions relying on vague or decentralized guidance [47, 76].”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [47] [47] Nora McDonald, Aditya Johri, Areej Ali, and Aayushi Hingle Collier. 2025. Generative artificial intelligence in higher education: Evidence from an analysis of institutional policies and guidelines. Computers in Human Behavior: Artificial Humans 3 (2025), 100121. - source

      The Introduction (p. 2) and section 4.8 (p. 18) document substantial variation in institutions’ GenAI guidance, from brief or minimal guidance to extensive policies. Section 4.1 (p. 9) says half left classroom approaches to instructors and that guidance on code and STEM activities was vague. Those findings do not establish that many institutions relied on vague or decentralized guidance overall; the paper also reports extensive guidance at many institutions. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites [76] [76] Hui Wang, Anh Dang, Zihao Wu, and Son Mac. 2024. Generative AI in higher education: Seeing ChatGPT through universities' policies, resources, and guidelines. Computers and Education: Artificial Intelligence 7 (2024), 100326. - source

  • Classified as probably not a substantial claim.

    Premisecitation
    § 3.2 ACAI: The Academic AI Capacity Index · p. 9

    Absent institutionalized mechanisms for participation, feedback, and revision, governance frameworks risk functioning as static or symbolic commitments rather than durable, contestable structures of authority.

    “Absent such mechanisms, governance frameworks risk functioning as static or symbolic commitments rather than durable, contestable structures of authority [49, 65].”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [49] [49] Jacob Metcalf, Emanuel Moss, Elizabeth Anne Watkins, Ranjit Singh, and Madeleine Clare Elish. 2021. Algorithmic impact assessments and accountability: The co-construction of impacts. In Proceedings of the 2021 ACM conference on fairness, accountability, and transparency . 735-746. FAccT '26, June 25-28, 2026, Montreal, QC, Canada - source
    • cites [65] [65] Andrew D Selbst. 2021. An institutional view of algorithmic impact assessments. Harv. JL & Tech. 35 (2021), 117. -

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation
    § C AI Detection Tools · p. 19

    Hadra et al. emphasize that for EFL-focused institutions, overreliance on machine-based identification risks misclassifying legitimate student work.

    “emphasizing that 'for [EFL]-focused institutions in particular, overreliance on machine-based identification risks misclassifying legitimate student work'”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [34] [34] Mohammad Hadra, Karleen Cambridge, and Mostefa Mesbah. 2026. Evaluating the accuracy and reliability of AI content detectors in academic contexts. International Journal for Educational Integrity 22, 1 (2026), 4. - source

  • May combine two claims.

    Premisecitation
    § C AI Detection Tools · p. 19

    Hadra et al. evaluated Turnitin and Originality across 192 texts and found overall accuracy of 0.61 and 0.69, respectively.

    “Hadra et al. [34] evaluated Turnitin and Originality across 192 texts - including authentic writing by English as a Foreign Language (EFL) students produced before the generative AI storm - and found overall accuracy of just 0.61 and 0.69, respectively”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [34] [34] Mohammad Hadra, Karleen Cambridge, and Mostefa Mesbah. 2026. Evaluating the accuracy and reliability of AI content detectors in academic contexts. International Journal for Educational Integrity 22, 1 (2026), 4. - source

  • May combine two claims.

    Premisecitation
    § 4 Findings · p. 13

    Wang et al. found that while 57% of universities in their dataset mentioned common tools, none explicitly recommended their use.

    “Wang et al. [76], who found that while 57% of universities in their dataset mentioned common tools, none explicitly recommended their use.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [76] [76] Hui Wang, Anh Dang, Zihao Wu, and Son Mac. 2024. Generative AI in higher education: Seeing ChatGPT through universities' policies, resources, and guidelines. Computers and Education: Artificial Intelligence 7 (2024), 100326. - source

      Section 5.2.3 (p. 8) reports that 59 of 104 universities (56.7%, approximately 57%) discussed common AI-detection tools. It also says none supported instructors using those tools to evaluate academic integrity or determine plagiarism, but it does not establish that none recommended their use for any purpose. (LLM, conf=0.86)

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim.

    Premisecitation
    § C AI Detection Tools · p. 19

    Hadra et al. conclude that AI detection tools' limitations make them unsuitable as the sole basis for decisions regarding academic misconduct.

    “Hadra et al. [34] conclude that 'their limitations make them unsuitable as the sole basis for decisions regarding academic misconduct,'”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [34] [34] Mohammad Hadra, Karleen Cambridge, and Mostefa Mesbah. 2026. Evaluating the accuracy and reliability of AI content detectors in academic contexts. International Journal for Educational Integrity 22, 1 (2026), 4. - source

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation
    § C AI Detection Tools · p. 20

    Prior work has documented the unreliability of AI detection tools and called for caution in their deployment in educational settings.

    “Prior work has documented the unreliability of AI detection tools and called for caution in their deployment in educational settings [34, 43, 77].”
    Quote in the paperSubstantial claim3 of 3 sources checked
    • cites [34] [34] Mohammad Hadra, Karleen Cambridge, and Mostefa Mesbah. 2026. Evaluating the accuracy and reliability of AI content detectors in academic contexts. International Journal for Educational Integrity 22, 1 (2026), 4. - source
    • cites [43] [43] Weixin Liang, Mert Yuksekgonul, Yining Mao, Eric Wu, and James Zou. 2023. GPT detectors are biased against non-native English writers. Patterns 4, 7 (2023), 100779. doi:10.1016/j.patter.2023.100779 - source
    • cites [77] [77] Debora Weber-Wulff, Alla Anohina-Naumeca, Sonja Bjelobaba, Tomáš Folt` ynek, Jean Guerrero-Dib, Olumide Popoola, Petr Šigut, and Lorna Waddington. 2023. Testing of detection tools for AI-generated text. International Journal for Educational Integrity 19, 1 (2023), 1-39. - source

  • May combine two claims.

    Premisecitation
    § C AI Detection Tools · p. 19

    Liang et al. found that seven widely used detectors misclassified over 47% of essays by non-native English speakers as AI-generated, while maintaining less than 12% misclassification rates for native English speakers.

    “Liang et al. [43] found that seven widely used detectors misclassified over 47% of essays by non-native English speakers as AI-generated, while maintaining less than 12% misclassifications rates for native English speakers.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [43] [43] Weixin Liang, Mert Yuksekgonul, Yining Mao, Eric Wu, and James Zou. 2023. GPT detectors are biased against non-native English writers. Patterns 4, 7 (2023), 100779. doi:10.1016/j.patter.2023.100779 - source

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Premisecitation
    § E.4 The Translation Workforce · p. 27

    Algorithmic action is inherently situated - systems can have unintended outcomes when deployed in real social and organizational contexts.

    “Algorithmic action is inherently situated - systems can have unintended outcomes when deployed in real social and organizational contexts [40, 67].”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites [40] [40] Michael Katell, Meg Young, Dharma Dailey, Bernease Herman, Vivian Guetler, Aaron Tam, Corinne Bintz, Daniella Raz, and P. M. Krafft. 2020. Toward situated interventions for algorithmic equity: lessons from the field. In Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency (Barcelona, Spain) (FAT* '20) . Association for Computing Machinery, New York, NY, USA, 45-55. doi:10.1145/3351095.3372874 - source
    • cites [67] [67] Lucy A. Suchman. 2007. Human-Machine Reconfigurations: Plans and Situated Actions (2nd ed.). Cambridge University Press, Cambridge, UK. - source

  • Probably not a substantial claim: may not pass S2 (checkable).

    Premisecitation
    § C AI Detection Tools · p. 22

    The usefulness in certain contexts and ubiquity of generative AI virtually guarantee its continued presence in certain academic and professional contexts.

    “At the same time, the usefulness in certain contexts and ubiquity of generative AI virtually guarantee its continued presence in certain academic and professional contexts [6].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [6] [6] Clare Baek, Tamara Tate, and Mark Warschauer. 2024. 'ChatGPT seems too good to be true': College students' use and perceptions of generative AI. Computers and Education: Artificial Intelligence 7 (2024), 100294. - source

  • May combine two claims.

    Findingown data
    § The Due Process Deficit: Auditing AI Governance in U.S. Higher Education · p. 1

    U.S. universities have adopted AI policies (revised honor codes, AI use guidelines, and detection tools), but the procedural infrastructure that would let affected parties understand, contest, and seek redress for AI-related decisions has not kept pace.

    “U.S. universities have met Artificial Intelligence (AI) with policy - revised honor codes, AI use guidelines, and detection tools to enforce them - but the procedural infrastructure that would let affected parties understand, contest, and seek redress for AI-related decisions has not kept pace.”
    • Also stated in: § 1 Introduction · p. 2 · plain
      “U.S. universities have met Artificial Intelligence (AI) with policy: revised honor codes, AI use guidelines, and detection tools to enforce them. What has not kept pace is the infrastructure needed to ensure that affected parties can understand, contest, and seek redress for decisions impacting them.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The 79 universities selected are sufficiently representative of U.S. higher education for the audit's findings to generalize to U.S. universities.

  • May combine two claims.

    Findingown data
    § The Due Process Deficit: Auditing AI Governance in U.S. Higher Education · p. 1

    Governance capacity is systematically concentrated in rule articulation (policy documents and faculty-facing guidance) while mechanisms for participation, feedback, and procedural constraint are underdeveloped.

    “We find that governance capacity is systematically concentrated in rule articulation while mechanisms for participation, feedback, and procedural constraint are underdeveloped”
    • Also stated in: § 1 Introduction · p. 3 · boosted
      “Ourfindings reveal that governance capacity is consistently concentrated in rule articulation - policy documents and faculty-facing guidance - while the mechanisms that enable participation, feedback, and procedural constraint are systematically underdeveloped.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The 79 universities selected are sufficiently representative of U.S. higher education for the audit's findings to generalize to U.S. universities.

  • Probably not a substantial claim: may not pass S4 (based).

    Findingnone given
    § 1 Introduction · p. 2

    Faculty are frequently expected to make consequential decisions about AI use without adequate institutional support, resources, or training.

    “Faculty, meanwhile, are frequently expected to make consequential decisions about AI use - whether and how to permit it, how to detect misuse, how to adjudicate ambiguous cases - without adequate institutional support, resources, or training.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The 79 universities selected are sufficiently representative of U.S. higher education for the audit's findings to generalize to U.S. universities.

  • May combine two claims.

    Findingnone given
    § 4 Findings · p. 12

    UC Berkeley is the highest-scoring institution among Large, Public Research, R1 universities in the West and achieves an ACAI score of 69.70.

    “UC Berkeley the highest-scoring institution among Large, Public Research, R1 universities in the West - achieves an ACAI score of 69.70.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 3.2 ACAI: The Academic AI Capacity Index · p. 11

    AI governance capacity scales with institutional type, research activity, and size.

    “AI governance capacity scales with institutional type, research activity, and size”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 4 Findings · p. 13

    The highest ACAI score is achieved by the University of New Hampshire, followed by Portland State University and Stanford University.

    “The highest ACAI score is achieved by the University of New Hampshire, followed by Portland State University and Stanford University.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingreasoning
    § C AI Detection Tools · p. 21

    Universities have frequently responded to the rise of generative tools with restrictive policies or punitive enforcement regimes, but these approaches are often built on vague or inconsistent definitions of permissible use.

    “Universities have frequently responded to the rise of generative tools with restrictive policies or punitive enforcement regimes, but these approaches are often built on vague or inconsistent definitions of permissible use.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 4 Findings · p. 14

    The errors are not symmetric: institutions in the top quartile of the rank-gap distribution experience misrankings of up to 64-72 positions, nearly inverting their relative ordering, while the bottom quartile shows discrepancies of 0-8 positions.

    “The errors are not symmetric: institutions in the top quartile of the rank-gap distribution experience misrankings of up to 64-72 positions - nearly inverting their relative ordering - while the bottom quartile shows nontrivial but modest discrepancies of 0-8 positions.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Findingown data
    § C AI Detection Tools · p. 20

    Our audit of 79 university AI policies extends prior work by examining whether calls for caution have been reflected in institutional practice.

    “Our audit of 79 university AI policies extends this line of work by examining whether these calls have been reflected in institutional practice.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 4 Findings · p. 13

    Although AI detection tools are frequently referenced in academic integrity materials, explicit procedural guidance governing their use is rare.

    “although AI detection tools are frequently referenced in academic integrity materials, explicit procedural guidance governing their use is rare.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 4 Findings · p. 14

    Across all temperatures, LLM-generated rankings exhibit moderate to strong ordinal agreement with human judgments (rho=0.54-0.62), but individual institutions are frequently and severely misranked, with mean absolute rank errors of 15-17 positions out of 79.

    “Across all temperatures, LLM-generated rankings exhibit moderate to strong ordinal agreement with human judgments ( 𝜌 = 0 . 54 -0 . 62), but individual institutions are frequently and severely misranked, with mean absolute rank errors of 15-17 positions out of 79.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 4 Findings · p. 14

    Mean maximum rank shifts were 16.32 ranks, with an interquartile range of 15.00-22.5 ranks and a maximum of 46 ranks.

    “Mean maximum rank shifts were a modest 16.32 ranks, with an interquartile range of 15.00-22.5 ranks and a maximum of 46 ranks”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 4 Findings · p. 13

    Institutions often fail to specify whether detection tools are advisory or determinative, how results should be interpreted, what safeguards exist against error, or what recourse is available to affected students.

    “Institutions often fail to specify whether detection tools are advisory or determinative, how results should be interpreted, what safeguards exist against error, or what recourse is available to affected students.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Findingnone given
    § C AI Detection Tools · p. 20

    ACAI extends prior work by auditing a structurally diverse institutional sample and shifting the analytic frame from pedagogical guidance to procedural accountability.

    “ACAI extends prior work by auditing a structurally diverse institutional sample and shifting the analytic frame from pedagogical guidance to procedural accountability.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § C AI Detection Tools · p. 19

    Nova Southeastern University specifically directs faculty to ZeroGPT, describing it as a tool where you can easily detect whether text is human-written or AI/GPT generated and noting that the company claims 99% and higher accuracy rate.

    “As shown in Table 6, Nova Southeastern University specifically directs faculty to ZeroGPT, describing it as a tool where '[y]ou can easily detect whether text is human-written or AI/GPT generated' and noting that the 'company claims 99% and higher accuracy rate.'”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation
    § 2.1 AI in Higher Education · p. 4

    The detection tools deployed in response to AI-assisted plagiarism are neither accurate nor reliable.

    “Despite widespread faculty concern about AI-assisted plagiarism [58], the detection tools deployed in response are neither accurate nor reliable - further discussed in Appendix C.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [58] [58] Britt S. Paris, Cynthia Conti-Cook, Daniel Greene, Kyle M. L. Jones, Brian Justie, Matthew Kirschenbaum, Lisa Kresge, Emma May, Aiha Nguyen, Rebecca Reynolds, et al. 2025. Artificial Intelligence and Academic Professions. Academe 111, 3 (July 2025), 49-59. - source

  • May combine two claims.

    Premisecitation
    § 2.3 Legal Grounding · p. 5

    Public universities bear constitutional obligations under the Fourteenth Amendment; both public and private institutions are bound by their own published procedures and subject to judicial review.

    “Public universities bear constitutional obligations under the Fourteenth Amendment; both public and private institutions are bound by their own published procedures and subject to judicial review [62].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [62] [62] Marie T Reilly. 2015. Due process in public university discipline cases. Penn St. L. Rev. 120 (2015), 1001. -

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning
    § 3.2 ACAI: The Academic AI Capacity Index · p. 8

    ACAI should be interpreted as a lower bound on institutional capacity.

    “Because the index relies on publicly available materials, it should be interpreted as a lower bound on institutional capacity.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation
    § 2.1 AI in Higher Education · p. 4

    More than half of students want education on ethical AI use, and more than half want clearer, standardized policies governing when and how AI tools can be used.

    “Students themselves recognize the gap: more than half want education on ethical AI use, and more than half want clearer, standardized policies governing when and how AI tools can be used [28].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [28] [28] Colleen Flaherty. 2025. How AI Is Changing-Not 'Killing'-College. Inside Higher Ed (29 Aug. 2025). https://www.insidehighered.com/ news/students/academics/2025/08/29/survey-college-students-views-ai -

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 2.1 AI in Higher Education · p. 4

    78% of faculty reported increased cheating since generative AI tools became widely available.

    “while 78% reported increased cheating since generative AI tools became widely available [61].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [61] [61] Lee Rainie and C. Edward Watson. 2026. The AI Challenge: How College Faculty Members Assess the Present and Future . Technical Report. Elon University's Imagining the Digital Future Center and American Association of Colleges and Universities (AAC&U). https://imaginingthedigitalfuture.org/wp-content/uploads/2026/01/Elon-AACU-faculty-AI-survey-full-report-1-21-26.pdf -

      Not checked in the free run.

  • May combine two claims.

    Premisecitation
    § 3.2 ACAI: The Academic AI Capacity Index · p. 9

    In the absence of clear, publicly legible policy language, responsibility for interpreting acceptable AI use is often shifted to individual faculty or students, increasing the risk of inconsistent enforcement and inequitable outcomes.

    “In the absence of clear, publicly legible policy language, responsibility for interpreting acceptable AI use is often shifted to individual faculty or students, increasing the risk of inconsistent enforcement and inequitable outcomes [58].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [58] [58] Britt S. Paris, Cynthia Conti-Cook, Daniel Greene, Kyle M. L. Jones, Brian Justie, Matthew Kirschenbaum, Lisa Kresge, Emma May, Aiha Nguyen, Rebecca Reynolds, et al. 2025. Artificial Intelligence and Academic Professions. Academe 111, 3 (July 2025), 49-59. - source

  • Classified as probably not a substantial claim.

    Premisecitation
    § 3.2 ACAI: The Academic AI Capacity Index · p. 9

    A central finding of the AAUP report was a widespread concern, reported by 71% of respondents, about the disconnect between administrative decision-making on AI policy and meaningful faculty and student input.

    “A central finding was a widespread concern (reported by 71% of respondents) about the disconnect between administrative decision-making on AI policy and meaningful faculty and student input.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [58] [58] Britt S. Paris, Cynthia Conti-Cook, Daniel Greene, Kyle M. L. Jones, Brian Justie, Matthew Kirschenbaum, Lisa Kresge, Emma May, Aiha Nguyen, Rebecca Reynolds, et al. 2025. Artificial Intelligence and Academic Professions. Academe 111, 3 (July 2025), 49-59. - source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data
    § 3.1 ACAI-US79: A Benchmarking Dataset · p. 7

    Each university was independently annotated by three annotators.

    “Each university was independently annotated by three annotators, administered via Google Forms.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation
    § 4 Findings · p. 12

    Public universities commonly operate under statutory or regulatory accountability frameworks and maintain centralized administrative infrastructures, visible as institution-level policies, guidance pages, and standing committees.

    “public universities commonly operate under statutory or regulatory accountability frameworks and maintain centralized administrative infrastructures, which are visible in the form of institution-level policies, guidance pages, and standing committees [8, 62, 78].”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [8] [8] Robert Birnbaum and Paul J Edelson. 1989. How colleges work: The cybernetics of academic organization and leadership. - source
    • cites [62] [62] Marie T Reilly. 2015. Due process in public university discipline cases. Penn St. L. Rev. 120 (2015), 1001. -

      Not checked in the free run.

    • cites [78] [78] Karl E Weick. 2012. Educational organizations as loosely coupled systems. In The roots of logistics . Springer, 389-410. - source

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationreasoning
    § 3.1 ACAI-US79: A Benchmarking Dataset · p. 7

    Including transient communications would collapse institutionalization into communication and systematically overstate governance capacity.

    “Including them would collapse institutionalization into communication and systematically overstate governance capacity.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning
    § 4 Findings · p. 12

    Lack of public legibility introduces an accountability risk: reliance on informal guidance or word-of-mouth governance differentially advantages actors with greater institutional access while disadvantaging students and faculty who must rely on publicly accessible rules and procedures.

    “this lack of public legibility introduces a specific accountability risk: reliance on informal guidance or word-of-mouth governance differentially advantages actors with greater institutional access, while disadvantaging students and faculty who must rely on publicly accessible rules and procedures.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning
    § 3.1 ACAI-US79: A Benchmarking Dataset · p. 7

    Transient communications such as news articles, announcements, or blog posts do not establish durable roles, procedures, or accountability structures.

    “While such materials may signal institutional intent or activity, they do not establish durable roles, procedures, or accountability structures.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationreasoning
    § 3.2 ACAI: The Academic AI Capacity Index · p. 9

    Faculty support directly impacts students: clear guidance helps faculty communicate consistent expectations to students, and faculty committees and resources provide the foundation for student-facing policies in other domains.

    “Importantly, faculty support directly impacts students: clear guidance helps faculty communicate consistent expectations to students, while faculty committees and resources provide the foundation for student-facing policies articulated in the other domains.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductnone given
    § 4 Findings · p. 12

    ACAI does not assess research quality, ethical commitments, or internal decision-making processes; it captures whether AI-related governance mechanisms are publicly articulated and institutionally maintained.

    “ACAI does not assess research quality, ethical commitments, or internal decision-making processes. It captures whether AI-related governance mechanisms are publicly articulated and institutionally maintained.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Publicly articulated governance mechanisms are a valid indicator of institutional AI governance capacity.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning
    § 3.2 ACAI: The Academic AI Capacity Index · p. 9

    The presence of formal faculty support mechanisms signals that responsibility for AI governance is not delegated entirely to individual faculty, but is recognized as an institutional obligation requiring shared infrastructure and expertise.

    “The presence of formal faculty support mechanisms signals that responsibility for AI governance is not delegated entirely to individual faculty, but is instead recognized also as an institutional obligation requiring shared infrastructure and expertise.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductreasoning
    § 3.2 ACAI: The Academic AI Capacity Index · p. 8

    We intentionally aggregate indicator scores by averaging across annotators rather than resolving disagreement through expert adjudication or majority vote.

    “We intentionally aggregate indicator scores by averaging across annotators rather than resolving disagreement through expert adjudication or majority vote.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductown data
    § 3.1 ACAI-US79: A Benchmarking Dataset · p. 7

    Two computational PhD students independently annotated a subset of 10 universities using the same audit protocol to assess alignment between expert and Prolific annotations.

    “To assess alignment between expert and Prolific annotations, two computational PhD students independently annotated a subset of 10 universities using the same audit protocol.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductreasoning
    § 3.3 Approximating AI Research Activity with CSRankings 𝐴𝐼 · p. 11

    CSRankings only ranks research-active institutions by publication output; Teaching/Liberal Arts colleges and institutions without substantial computer science research activity do not appear in CSRankings and are therefore excluded from the correlation analysis in Figure 10.

    “Notably, CSRankings only ranks research-active institutions by publication output; Teaching/Liberal Arts colleges and institutions without substantial computer science research activity do not appear in CSRankings and are therefore excluded from the correlation analysis in Figure 10.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductreasoning
    § 3.1 ACAI-US79: A Benchmarking Dataset · p. 7

    Annotators were instructed not to use external search engines or prior knowledge, and each item was evaluated under a strict time limit.

    “To preserve the audit's focus on public legibility, annotators were instructed not to use external search engines or prior knowledge, and each item was evaluated under a strict time limit.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 3.2 ACAI: The Academic AI Capacity Index · p. 9

    The July 2025 AAUP report Artificial Intelligence and Academic Professions draws on approximately 500 survey responses from AAUP members regarding their experiences with AI and other educational technologies.

    “The report draws on approximately 500 survey responses from AAUP members regarding their experiences with AI and other educational technologies.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [58] [58] Britt S. Paris, Cynthia Conti-Cook, Daniel Greene, Kyle M. L. Jones, Brian Justie, Matthew Kirschenbaum, Lisa Kresge, Emma May, Aiha Nguyen, Rebecca Reynolds, et al. 2025. Artificial Intelligence and Academic Professions. Academe 111, 3 (July 2025), 49-59. - source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning
    § 3.1 ACAI-US79: A Benchmarking Dataset · p. 7

    Prolific's 'reject exceptionally fast submissions' safeguard was enabled to support data quality.

    “To support data quality, we enabled Prolific's 'reject exceptionally fast submissions' safeguard, which automatically flags submissions completed at implausibly short durations relative to the estimated task time, helping to filter inattentive or agent-driven responses.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning
    § 3.2 ACAI: The Academic AI Capacity Index · p. 8

    We assess robustness under alternative weighting schemes, varying 1 ≤ w_d ≤ 4.

    “To ensure findings are not artifacts of these choices, we assess robustness under alternative weighting schemes, varying 1 ≤ 𝑤 𝑑 ≤ 4 (see Finding 5 and Appendix H).”
    Quote in the paperNot a substantial claim

  • May combine two claims. The quote only partly states this claim.

    Study design and conductown data
    § 4 Findings · p. 14

    ACAI rankings are robust to weighting choices and individual annotators.

    “ACAI rankings are robust to weighting choices and individual annotators.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationnone given
    § 5 Discussion · p. 15

    Reliance on publicly available institutional data may omit informal practices or internal decision-making processes that shape outcomes but are not externally visible.

    “our reliance on publicly available institutional data may omit informal practices or internal decisionmaking processes that shape outcomes but are not externally visible.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning
    § 5 Discussion · p. 15

    Non-public materials may constitute genuine governance infrastructure, and their exclusion means ACAI should be interpreted as a lower bound on governance capacity.

    “Such materials may constitute genuine governance infrastructure, and their exclusion means ACAI should be interpreted as a lower bound on governance capacity.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data
    § 4 Findings · p. 14

    While LLMs may approximate coarse aggregate patterns, they fail to reliably reproduce the fine-grained, interpretive distinctions required for institutional governance audits.

    “while LLMs may approximate coarse aggregate patterns, they fail to reliably reproduce the fine-grained, interpretive distinctions required for institutional governance audits.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning
    § 4 Findings · p. 13

    The gap in procedural guidance for detection tools creates ambiguity around authority and enforcement and risks inconsistent or discretionary application in practice, directly harming students and creating potential for an adversarial relationship between students and faculty.

    “This gap creates ambiguity around authority and enforcement, and risks inconsistent or discretionary application in practice, as expanded upon in Appendix C, directly harming students, and creating the potential for an adversarial relationship between students and faculty which is counterproductive to a healthy learning environment.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation
    § C AI Detection Tools · p. 21

    The result of vague or inconsistent definitions is widespread uncertainty among both students and faculty, inconsistent application of standards, and in some cases, false accusations of academic misconduct.

    “The result is widespread uncertainty among both students and faculty [75], inconsistent application of standards, and in some cases, false accusations of academic misconduct.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [75] [75] James D. Walsh. 2025. Everyone Is Cheating Their Way Through College: ChatGPT has unraveled the entire academic project. Intelligencer (New York Magazine) (7 May 2025). https://nymag.com/intelligencer/article/openai-chatgpt-ai-cheating-education-college-studentsschool.html -

      Not checked in the free run.

  • May combine two claims.

    Study design and conductown data
    § 4 Findings · p. 14

    Across all three leave-one-annotator-out exclusions, recomputed ACAI scores remained strongly correlated with the full-annotator index (rho=0.77-0.87, r=0.83-0.87), indicating that no single annotator systematically altered the relative ordering of high- or low-capacity institutions.

    “Across all three exclusions, recomputed scores remained strongly correlated with the full-annotator index ( 𝜌 = 0.77-0.87, 𝑟 = 0.83-0.87), indicating that no single annotator systematically altered the relative ordering of high- or low-capacity institutions.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning
    § 4 Findings · p. 13

    Differences in AI governance capacity are primarily quantitative rather than qualitative: capacity scales with organizational resources, but its internal composition remains skewed toward static guidance rather than durable procedural safeguards.

    “differences are primarily quantitative rather than qualitative: AI governance capacity scales with organizational resources, but its internal composition remains skewed toward static guidance rather than durable procedural safeguards.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning
    § 5 Discussion · p. 15

    The analysis is USA-centric and may not generalize to other geographies.

    “the analysis is USA-centric, reflecting the regulatory, cultural, and policy context of U.S. higher education, but may not generalize to other geographies.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data
    § 4 Findings · p. 14

    When indicators are aggregated at the governance domain-level, agreement improves (alpha=0.30), suggesting that higher-level governance constructs are more consistently interpretable than individual policy statements.

    “When indicators are aggregated at the governance domain-level, agreement improves ( 𝛼 = 0 . 30), suggesting that higher-level governance constructs are more consistently interpretable than individual policy statements.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductreasoning
    § F ACAI-US79 Iterative Link Categorization Process · p. 28

    The dataset should be interpreted as a representation of officially articulated AI governance rather than a complete account of institutional practice.

    “As a result, the dataset should be interpreted as a representation of officially articulated AI governance rather than a complete account of institutional practice.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § C AI Detection Tools · p. 21

    Many faculty are concerned with academic dishonesty with the rise of generative AI.

    “While it is true that many faculty are concerned with academic dishonesty with the rise of generative AI”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [58] [58] Britt S. Paris, Cynthia Conti-Cook, Daniel Greene, Kyle M. L. Jones, Brian Justie, Matthew Kirschenbaum, Lisa Kresge, Emma May, Aiha Nguyen, Rebecca Reynolds, et al. 2025. Artificial Intelligence and Academic Professions. Academe 111, 3 (July 2025), 49-59. - source

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Study design and conductreasoning
    § F ACAI-US79 Iterative Link Categorization Process · p. 28

    By narrowing annotator attention to governance-relevant materials, Phase III also improves annotation consistency and strengthens the construct validity of the audit.

    “By narrowing annotator attention to governance-relevant materials, Phase III also improves annotation consistency and strengthens the construct validity of the audit.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductreasoning
    § D Public Website · p. 23

    The website is designed to make both the audit results and the audit instrument accessible beyond the paper, enabling readers to explore institutional AI governance capacity and to trace aggregate scores back to the underlying publicly available materials.

    “The site is designed to make both the audit results and the audit instrument accessible beyond the paper, enabling readers to explore institutional AI governance capacity and to trace aggregate scores back to the underlying publicly available materials.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given
    § J Details on LLM-Driven Audit Study · p. 34

    Outputs failing schema or completeness checks were automatically retried until a valid response was produced.

    “Outputs failing schema or completeness checks were automatically retried until a valid response was produced.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning
    § D Public Website · p. 23

    The website design explicitly supports traceability, allowing users to inspect how scores are grounded in publicly legible governance artifacts rather than treating the index as an opaque ranking.

    “This design explicitly supports traceability, allowing users to inspect how scores are grounded in publicly legible governance artifacts rather than treating the index as an opaque ranking.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data
    § CSRankings: Computer Science Rankings · p. 41

    The LLM robustness study used a prompt that mirrored the instructions given to human annotators in the ACAI audit.

    “Prompt used in the LLM robustness study. The figure shows the full evaluation prompt provided to the language model, including source restrictions, scoring criteria, and the required JSON output schema, mirroring the instructions given to human annotators in the ACAI audit.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning
    § C AI Detection Tools · p. 22

    Preventing misuse is not simply a question of enforcement but one of governance design.

    “These challenges highlight that preventing misuse is not simply a question of enforcement but one of governance design.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductreasoning
    § F ACAI-US79 Iterative Link Categorization Process · p. 28

    Phase III enforces the conceptual boundary of the audit by removing links that are AI-related but governance-irrelevant. This consolidation step prevents research-intensive institutions from appearing more 'governed' simply due to higher volumes of AI-related content and ensures that institutional capacity is not conflated with research productivity.

    “Phase III enforces the conceptual boundary of the audit by removing links that are AI-related but governance-irrelevant, such as research labs, grants, or faculty-led initiatives that do not articulate institutional authority or procedural expectations. This consolidation step prevents research-intensive institutions from appearing more 'governed' simply due to higher volumes of AI-related content”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § C AI Detection Tools · p. 21

    For at least some faculty, the concern goes beyond the dishonesty question and is rooted in a worry that students will fail to learn.

    “it is also the case that for at least some faculty, the concern goes beyond the (dis)honesty question and is rooted in a worry that students will fail to learn”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [58] [58] Britt S. Paris, Cynthia Conti-Cook, Daniel Greene, Kyle M. L. Jones, Brian Justie, Matthew Kirschenbaum, Lisa Kresge, Emma May, Aiha Nguyen, Rebecca Reynolds, et al. 2025. Artificial Intelligence and Academic Professions. Academe 111, 3 (July 2025), 49-59. - source

  • May combine two claims.

    Interpretationcitation
    § C AI Detection Tools · p. 22

    Even a 1% false positive rate can result in thousands of wrongful accusations across large institutions, eroding trust in both faculty and administrative oversight.

    “Even a 1% false positive rate can result in thousands of wrongful accusations across large institutions, eroding trust in both faculty and administrative oversight, as illustrated in Figure 4.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [29] [29] University Center for Teaching and University of Pittsburgh Learning. 2025. Encouraging Academic Integrity. Online resource. https://teaching.pitt.edu/resources/encouraging-academic-integrity/ -

      Not checked in the free run.

    • cites [71] [71] Turnitin. 2023. Understanding false positives within our AI writing detection capabilities. https://www.turnitin.com/blog/understandingfalse-positives-within-our-ai-writing-detection-capabilities. -

      Not checked in the free run.

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductnone given
    § D Public Website · p. 24

    ACAI scores reflect the public legibility of institutional governance artifacts at the time of review and should not be interpreted as measures of internal practice or intent.

    “ACAI scores reflect the public legibility of institutional governance artifacts at the time of review and should not be interpreted as measures of internal practice or intent.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Study design and conductnone given
    § I Details on Comparison to CSRankings · p. 33

    This procedure relies exclusively on publicly accessible filters and settings within CSRankings and is fully reproducible.

    “This procedure relies exclusively on publicly accessible filters and settings within CSRankings and is fully reproducible.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: CSRankings is a widely used, publicly available ranking of computer science research output.
    • Rests on an assumption: CSRankings aggregates publication counts across major computer science venues and allows filtering by research area and time period.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning
    § E.4 The Translation Workforce · p. 27

    Effective governance therefore requires people and processes capable of translating between computational reasoning and institutional judgment.

    “Effective governance therefore requires people and processes capable of translating between computational reasoning and institutional judgment.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductreasoning
    § D Public Website · p. 24

    By making both the audit results and the audit instrument publicly accessible, the site supports replication, critique, and institutional self-reflection, and positions ACAI as an auditable, contestable infrastructure rather than a static evaluation.

    “By making both the audit results and the audit instrument publicly accessible, the site supports replication, critique, and institutional self-reflection, and positions ACAI as an auditable, contestable infrastructure rather than a static evaluation.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning
    § C AI Detection Tools · p. 23

    At scale in a university setting, for every 100,000 submissions, 1000 false accusations could occur, jeopardizing student careers and significantly de-legitimizing educational institutions.

    “At scale in a university setting, for every 100,000 submissions, 1000 false accusations could occur, jeopardizing student careers and significantly de-legitimizing educational institutions.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [29] [29] University Center for Teaching and University of Pittsburgh Learning. 2025. Encouraging Academic Integrity. Online resource. https://teaching.pitt.edu/resources/encouraging-academic-integrity/ -

      Not checked in the free run.

    • cites [71] [71] Turnitin. 2023. Understanding false positives within our AI writing detection capabilities. https://www.turnitin.com/blog/understandingfalse-positives-within-our-ai-writing-detection-capabilities. -

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning
    § C AI Detection Tools · p. 22

    Institutions that tolerate high rates of false positives in AI detection risk delegitimizing their authority and weakening the educational contract.

    “Institutions that tolerate high rates of false positives in AI detection - for example, by punishing students based on unreliable systems - risk delegitimizing their authority and weakening the educational contract itself.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Study design and conductreasoning
    § F ACAI-US79 Iterative Link Categorization Process · p. 27

    Because institutional AI governance is unevenly distributed and inconsistently labeled across universities, we employ a recall-oriented, multi-phase procedure that surfaces publicly legible, institutionally authoritative materials while enforcing a clear boundary between governance capacity and general AI research or outreach.

    “Because institutional AI governance is unevenly distributed and inconsistently labeled across universities, we employ a recall-oriented, multi-phase procedure that surfaces publicly legible, institutionally authoritative materials while enforcing a clear boundary between governance capacity and general AI research or outreach.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given
    § D Public Website · p. 23

    The self-scoring tool implements the same annotation schema and scoring criteria used in the audit, enabling users to assess their own institution using publicly available materials under comparable constraints.

    “This tool implements the same annotation schema and scoring criteria used in the audit, enabling users to assess their own institution using publicly available materials under comparable constraints.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § E.1 Institutional & Socio-Technical Systems Theory, and Algorithmic Accountability · p. 24

    Technological and institutional capacities do not evolve independently, instead co-producing and mutually constructing each other.

    “Technological and institutional capacities do not evolve independently, instead co-producing and mutually constructing each other [38].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [38] [38] Sheila Jasanoff et al. 2004. States of knowledge . Taylor & Francis Abingdon, UK. -

      Not checked in the free run.

  • May combine two claims. The quote only partly states this claim.

    Interpretationreasoning
    § C AI Detection Tools · p. 21

    False accusations and inconsistent application of standards erode trust and undermine institutional legitimacy.

    “Such outcomes erode trust, undermine institutional legitimacy”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning
    § C AI Detection Tools · p. 22

    Attempts to suppress use via detection technologies are both impractical and counterproductive.

    “Attempts to suppress use via detection technologies are both impractical and counterproductive.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning
    § F.1 Phase II vs. III Coverage Comparison · p. 30

    Human supervision is necessary to filter non-authoritative links, recover missed policy documents, and ensure that final policy representations accurately reflect institutional governance.

    “However, the observed gap between apparent and aligned coverage underscores the continued necessity of human supervision to filter non-authoritative links, recover missed policy documents, and ensure that final policy representations accurately reflect institutional governance.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Study design and conductnone given
    § F ACAI-US79 Iterative Link Categorization Process · p. 28

    Universities differ substantially in how AI-related guidance is organized, labeled, and distributed across administrative units, which may lead to uneven coverage across institutions.

    “First, universities differ substantially in how AI-related guidance is organized, labeled, and distributed across administrative units, which may lead to uneven coverage across institutions.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductreasoning
    § F ACAI-US79 Iterative Link Categorization Process · p. 28

    The dataset captures only formal, publicly visible governance artifacts and does not reflect informal practices, internal guidance, or unpublished decision-making processes.

    “Second, the dataset captures only formal, publicly visible governance artifacts and does not reflect informal practices, internal guidance, or unpublished decision-making processes that may substantially shape how AI is used and regulated within institutions.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductreasoning
    § F ACAI-US79 Iterative Link Categorization Process · p. 27

    During manual screening, we remove dead links, incorrect links, AI-hallucinated links, news links, and announcement pages. This screening step ensures that the resulting corpus contains only live, correctly-attributed, on-scope institutional pages suitable for downstream analysis.

    “During manual screening, we remove dead links, incorrect links (e.g., pages from the wrong university), AI-hallucinated links that do not exist, news links, and announcement pages. This screening step ensures that the resulting corpus contains only live, correctly-attributed, on-scope institutional pages suitable for downstream analysis.”
    Quote in the paperSubstantial claim
89 of 89

  • The paper relies on this without support.

    Universities exercise significant power over students and faculty.

    “Universities exercise significant power over students and faculty, 1 and AI-related governance decisions carry material consequences.”

  • The paper relies on this without support.

    The 79 universities selected are sufficiently representative of U.S. higher education for the audit's findings to generalize to U.S. universities.

    “The dataset was constructed using a purposive, diversity-oriented selection strategy designed to surface variation in how AI governance is institutionalized across U.S. higher education.”

  • The paper relies on this without support.

    Publicly articulated governance mechanisms are a valid indicator of institutional AI governance capacity.

    “It captures whether AI-related governance mechanisms are publicly articulated and institutionally maintained.”

  • The paper relies on this without support.

    The ACAI scoring domains (Policy Clarity, Faculty Support, Feedback Mechanisms, Detection Tools) adequately capture the construct of AI governance capacity.

    “scores in A. Policy Clarity and B. Faculty Support are systematically higher than those in C. Feedback Mechanisms and D. Detection Tools.”

  • The paper relies on this without support.

    CSRankings AI is a valid measure of AI-specific research output.

    “AI governance capacity (ACAI) is weakly correlated with AI-specific research output (CSRankings 𝐴𝐼 )”

  • The paper relies on this without support.

    Faculty are ill-equipped to assess state-of-the-art AI systems.

    “It falls on the instructor to evaluate use - faculty who are ill-equipped to assess state-of-the-art AI systems.”

  • The paper relies on this without support.

    Computational capacity's rapid expansion has outpaced the institutional systems needed for accountability.

    “While computational capacity has been extensively benchmarked, its rapid expansion has outpaced the institutional systems needed for accountability.”

  • The paper relies on this without support.

    Concerns about plagiarism and unauthorized use of generative AI tools are often treated as isolated classroom management issues.

    “Concerns about plagiarism and unauthorized use of generative AI tools are often treated as isolated classroom management issues.”

  • The paper relies on this without support.

    CSRankings is a widely used, publicly available ranking of computer science research output.

    “CSRankings, shown in Figure 11, a widely used, publicly available ranking of computer science research output.”

  • The paper relies on this without support.

    CSRankings aggregates publication counts across major computer science venues and allows filtering by research area and time period.

    “CSRankings aggregates publication counts across major computer science venues and allows filtering by research area and time period.”

  • The paper relies on this without support.

    The period 2022-2025 corresponds to the period following the public release of large-scale generative AI systems (e.g., ChatGPT) and the rapid expansion of AI use in educational contexts.

    “We then restricted the publication window to 2022-2025, corresponding to the period following the public release of large-scale generative AI systems (i.e., ChatGPT) and the following rapid expansion of AI use in educational contexts.”

  • The paper relies on this without support.

    The period 2022-2025 reflects contemporary AI research activity during the period in which universities began articulating institutional responses to generative AI.

    “This temporal filter was chosen to reflect contemporary AI research activity during the period in which universities began articulating institutional responses to generative AI.”
12 of 12

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Leave-one-annotator-out weighting scheme is inconsistent

    Minor

    Appendix H says the leave-one-out analysis uses the equal-weighted score, while Table 9 identifies the same analysis as using indicator-weighted ACAI scores.

  • Reported leave-one-out correlation range excludes a table value

    Minor

    The text gives an upper Pearson correlation of 0.87, but Table 9 reports 0.880 for the third annotator exclusion, which rounds to 0.88 at the text's precision. Its reported Spearman upper bound of 0.87 likewise excludes the table's 0.878, which rounds to 0.88.

2 of 2

Cites

Flagged first, then by citations

Research works printed but not linked: 20

  1. Dean W. Ball and Ketan Ramakrishnan. 2025. Entity-Based Regulation in Frontier AI Governance. Carnegie Endowment for International Peace (2025). https://carnegieendowment.org/research/2025/06/artificial-intelligence-regulation-united-states?lang=en Published July 7, 2025.
  2. William Blackstone. 1765. Commentaries on the Laws of England. Clarendon Press, Oxford. Book 4, Chapter 27 has the quote..
  3. Marcus Alphons Petrus Bovens Robert E Goodin and Thomas Schillemans. 2014. The Oxford handbook of public accountability. Oxford handbooks.
  4. Simon Coghlan, Tim Miller, and Jeannie Paterson. 2021. Good proctor or “big brother”? Ethics of online exam supervision technologies. Philosophy & Technology 34, 4 (2021), 1581–1606.
  5. Glenn Colby. 2025. Data Snapshot: Tenure and Contingency in US Higher Education, Fall 2023. Academe 111, 2 (Spring 2025). https://www.aaup.org/academe/issues/spring-2025/data-snapshot-tenure-and-contingency-us-higher-education-fall-2023
  6. Ethan Tuttle. 2026. LSU Students Face Mounting AI Cheating Allegations; Detection Technology Creates Campus Controversy. WAFB Channel 9 News. https://www.wafb.com/2026/01/14/lsu-students-face-mounting-ai-cheating-allegations-detection-technology-creates-campus-controversy/
  7. Colleen Flaherty. 2025. How AI Is Changing—Not ‘Killing‘—College. Inside Higher Ed (29 Aug. 2025). https://www.insidehighered.com/news/students/academics/2025/08/29/survey-college-students-views-ai
  8. Lon Luvois Fuller. 1964. The Morality of Law. Yale University Press, New Haven, CT.
  9. Margot E. Kaminski and Jennifer M. Urban. 2021. The Right to Contest AI. Columbia Law Review 121, 7 (2021). https://www.columbialawreview.org/content/the-right-to-contest-ai/
  10. Stefanie A Lindquist and Frank C Cross. 2010. Stability, predictability and the rule of law: stare decisis as reciprocity norm. In Conference on Legal Decision Making (Indiana).
  11. Florence Martin, Drew Polly, Shanna Coles, and Chuang Wang. 2020. Examining higher education faculty use of current digital technologies: Importance, competence, and motivation. International Journal of Teaching and Learning in Higher Education 32, 1 (2020), 73–86.
  12. Alexi Mostrous, Serena Cesareo, and Joe White. 2024. The Global Artificial Intelligence Index 2024. Tortoise Media (Sep 2024). https://www.tortoisemedia.com/2024/09/18/the-global-artificial-intelligence-index-2024
  13. Pablo Fuentes Nettel, Emma Hankins, Richard Stirling, Giulia Cirri, Gonzalo Grau, Sulamaan Rahim, and Eddie Crampton. 2024. Government AI Readiness Index 2024. Report 7th Edition. Oxford Insights.
  14. Jessica Priest. 2025. Texas universities deploy AI tools to review and rewrite how some courses discuss race and gender. The Texas Tribune (15 Dec. 2025). https://www.texastribune.org/2025/12/15/texas-universities-ai-course-audits/
  15. Marie T Reilly. 2015. Due process in public university discipline cases. Penn St. L. Rev. 120 (2015), 1001.
  16. Teresa Scassa. 2022. The surveillant university: Remote proctoring, AI, and human rights. Can. J. Comp. & Contemp. L. 8 (2022), 271.
  17. Andrew D Selbst. 2021. An institutional view of algorithmic impact assessments. Harv. JL & Tech. 35 (2021), 117.
  18. James D. Walsh. 2025. Everyone Is Cheating Their Way Through College: ChatGPT has unraveled the entire academic project. Intelligencer (New York Magazine) (7 May 2025). https://nymag.com/intelligencer/article/openai-chatgpt-ai-cheating-education-college-students-school.html
  19. Langdon Winner. 1980. Do Artifacts Have Politics? Daedalus (1980), 121–136.
  20. Jiawei Zhang and Boris Paal. 2025. A Three-Layered Framework: An AI Governance Guide for Global Policymakers. (2025). https://papers.ssrn.com/abstract=5241351

Cited by

Sorted by citations

Authors
Teleki, M., Choi, A. S. G., Duray, A., Liu, H., Zhang, J., Dong, X., Da Silva, D., Koenecke, A., & Caverlee, J.
Affiliations
  • Texas A&M University (US)
  • Cornell University (US)
  • American Association For The Advancement of Science (US)
  • Independent, College Station, USA
  • Texas A&M University, College Station, USA
  • Cornell Tech (US)
ORCID
Venue
Proceedings of the 2026 ACM Conference on Fairness, Accountability, and Transparency
Publisher
ACM
Published
2026-06-25
Pages
3738-3778
DOI
10.1145/3805689.3812328
OpenAlex
W7166564305
Semantic Scholar
1102a4e6db0c6f94ea75c84fea7fa713e3a9efff
Cited by
0
References
42
FWCI
0.00
Type
conference paper
Open access
Licence
Locations
ACM Conference on Fairness, Accountability, and Transparency (FAccT) (publishedVersion, cc-by)
Topics
  • Social Sciences › Safety Research › Ethics and Social Impacts of AI
  • Social Sciences › Law › Legal Issues in Education
  • Physical Sciences › Computer Science › Information Systems › Artificial Intelligence in Education
Keywords
AI governance, due process, higher education, algorithmic decision-making, algorithmic transparency, student rights

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-25.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-25.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review