Pith. sign in

REVIEW 4 major objections 5 minor 36 references

A two-stage framework claims that bias governance in skills-based job matching must span skill extraction and multistakeholder recommendation, linked by hard and soft constraints.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review

2026-08-01 22:31 UTC pith:T5YOUCOU

load-bearing objection A coherent, honestly-scoped framework paper that integrates known fairness methods into a two-stage AI Act governance architecture — but the load-bearing threshold mechanism is described, not operationalized. the 4 major comments →

arxiv 2607.15707 v1 pith:T5YOUCOU submitted 2026-07-17 cs.IR

From Skill Extraction to Multistakeholder Recommendation: A Two-Stage Framework for Bias Governance in Skills-Based Job Matching

Andrea Forster , Gregor Autischer , Dominik Kowald , Simone Kopeinik This is my paper
classification cs.IR
keywords bias governanceskill extractionjob matchingmultistakeholder recommendationsocial choiceEU AI Actcounterfactual testingfairness thresholds
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper is trying to establish that bias in AI-driven job platforms cannot be governed only at the ranking step: errors and biases introduced when candidate skills are collected and structured can propagate into recommendations. It proposes a two-stage framework in which Stage 1 audits skill extraction via distributional auditing and counterfactual testing, and Stage 2 runs candidate-job recommendation as three stakeholder agents (candidate, company, regulator) whose rankings are combined by social-choice voting. A shared hard/soft constraint mechanism connects the stages: hard violations block progression until corrected, while soft findings are logged and carried forward as fairness thresholds for the regulator agent. If the framework is right, a job platform can detect extraction bias before it reaches the recommender, make stakeholder trade-offs explicit and auditable, and maintain a compliance trail for the EU AI Act. The paper presents this as a conceptual architecture, not an evaluated system.

Core claim

The paper's central claim is that bias governance for skills-based job matching should be organized as a two-stage process under one governance layer. In Stage 1, skill extraction and profile formation are audited using historical distributional analysis and counterfactual tests, producing a bias inventory of hard constraints that block a profile and soft constraints that are passed forward. In Stage 2, candidate, company, and regulator objectives are represented as separate agents, each producing an independent ranking, and these rankings are aggregated through social-choice-based voting into a single auditable recommendation. The same hard/soft logic governs the recommendation: fairness me

What carries the argument

The central mechanism is the hard/soft constraint vocabulary applied at both stages: hard constraints are violations that block downstream processing until corrected, and soft constraints are tolerable findings that are logged and carried forward. In Stage 1 the constraints come from a bias inventory built by distributional auditing (comparing skill-assignment shares across demographic groups) and counterfactual testing (gender expression substitution and stereotype injection). In Stage 2 the constraints become fairness thresholds for a Regulator agent, and the carrying object is the set of stakeholder rankings (Candidate, Company, Regulator) combined by social-choice aggregation rules, such

Load-bearing premise

The framework's gates and re-triggering depend on 'predefined fairness thresholds' and on distributional auditing producing reliable demographic baselines, yet the paper offers no procedure for choosing those thresholds and concedes (Section 4) that they remain design decisions requiring empirical validation.

What would settle it

Run the Stage 1 counterfactual tests (gender expression substitution, stereotype injection) on a production chatbot-based skill extractor and then run the Stage 2 aggregation with and without the Regulator agent on the same candidate-job pool. If the extracted profiles do not shift under counterfactual edits, or if including the Regulator agent never changes the top-k recommendations, the framework's claimed bias-detection and governance effects are not observable in practice.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Extraction-stage biases can be caught and corrected before they enter the recommender, rather than only being mitigated at ranking time.
  • Trade-offs among candidate, company, and regulator objectives become explicit because each is represented by a separate ranking and the aggregation rule is documented.
  • Fairness is treated as a long-term property: soft-constraint findings accumulate into persistent fairness states, enabling detection of gradual drift.
  • The hard/soft gate gives a concrete audit trail that can support preparation for EU AI Act compliance for high-risk employment AI.
  • The framework implies that the choice of voting rule and agent weights is itself a fairness decision, since different aggregation rules yield different outcomes.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same hard/soft governance pattern could be lifted out of job matching into other high-stakes, multi-stakeholder AI pipelines—credit scoring, education, or public services—where input construction and downstream ranking both shape access.
  • The framework's reliance on historical distributional baselines means that thresholds derived naively could entrench historical segregation; a natural test is to compare threshold-setting procedures for sensitivity to baseline choice before deployment.
  • If Stage 1 counterfactual testing is run on current LLM-based skill extractors, it should reveal measurable divergence under gender expression substitution; that result would empirically motivate the framework's central premise.
  • Social-choice aggregation suggests a concrete research agenda: measuring which voting rules best preserve each agent's fairness objective while maintaining recommendation relevance; the paper leaves this open.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes a two-stage conceptual framework for bias detection and governance in skills-based job matching. Stage 1 covers candidate skill extraction and profile formation (with emphasis on chatbot-based elicitation) and uses distributional auditing and counterfactual testing to produce a bias inventory of hard and soft constraints. Stage 2 is a multistakeholder recommender in which Candidate, Company, and Regulator agents each produce rankings that are aggregated by social-choice voting; post-recommendation evaluation applies the same hard/soft logic, retriggering adapted configurations on threshold violations and logging soft findings as fairness states. A structured assessment based on the Fraunhofer AI Assessment Catalog is proposed for EU AI Act auditability. Sections 4 and 5 explicitly state that the framework is conceptual and has not been evaluated as a complete system.

Significance. If instantiated and validated, the framework would provide a useful architectural synthesis: it connects pre-ranking extraction bias to downstream recommendation governance, makes stakeholder trade-offs explicit via agent rankings and social-choice aggregation, and gives a concrete compliance-preparation route under the EU AI Act. Strengths include the clear mapping of audit methods (distributional auditing, gender-expression substitution, stereotype injection), the explicit discussion of voting-rule dependence, and an unusually honest limitation statement. However, the paper contains no implementation, data, or formal model, and the key operating parameters (thresholds, weights, aggregation rules) remain unspecified. Its current value is as a research agenda or reference architecture rather than a demonstrated framework.

major comments (4)
  1. [3.2.2, 3.4] The hard/soft gates and the Regulator agent rest on 'predefined fairness thresholds' said to be 'derived from the historical bias audit and counterfactual tests,' yet no procedure maps an audit output or counterfactual divergence to a numerical threshold, and no correctness criterion is given. Section 4 concedes that thresholds are design decisions requiring empirical validation. As written, the framework cannot be instantiated or tested: a miscalibrated threshold either lets discriminatory profiles through the hard gate or blocks qualified candidates. Specify a threshold-selection procedure and an evaluation criterion (e.g., error rates on simulated ground-truth profiles) before the gate mechanism can be considered operational.
  2. [3.1.2] Distributional auditing proposes flagging skill clusters where assignment shares 'deviate substantially from the overall population distribution' or 'historical proportions.' This makes historical/observed proportions the normative baseline, but historical data may encode the very discrimination the framework aims to govern. The manuscript never distinguishes a descriptive baseline from a normative fairness target, so the proposed regulator could preserve historical skew while appearing balanced. Add an explicit normative target or a correction for baseline bias; otherwise the audit's output is ambiguous.
  3. [3.4] The post-recommendation loop tests 'alternative agent configurations' and 'aggregation rules' when a threshold is crossed, then re-evaluates with the same metrics, but provides no stopping rule, convergence criterion, or method for selecting among configurations that all pass the threshold. The governance loop is therefore underdetermined: it can cycle indefinitely or settle arbitrarily. Specify a decision rule (e.g., lexicographic optimization over stakeholder objectives, or a budgeted search) and define what counts as a resolved hard violation.
  4. [Abstract, 4, 5] The manuscript repeatedly and correctly states that the framework is unvalidated. Because the central claim is architectural rather than empirical, the absence of experiments alone is not disqualifying for a conceptual paper. However, the abstract's 'the result is a framework ... into a single governance layer' overstates the status of an unimplemented proposal. Recommend framing this as a proposed architecture, and ideally include a minimal worked example or synthetic simulation showing that Stage 1 audit outputs can, in principle, determine Stage 2 thresholds.
minor comments (5)
  1. [2.3.3] Heading contains an erroneous space: 'F raunhofer AI Assessment Catalog' should be 'Fraunhofer AI Assessment Catalog'.
  2. [Figure 1] Figure 1 is referenced in Section 3, but the full text shows only the caption and no actual figure. A rendered figure or a more precise reference to the caption placeholder is needed.
  3. [3.3.3] The paper cites [26] for demographic parity assessed via Kullback-Leibler or Jensen-Shannon divergence, but [26] analyzes item popularity bias in music recommenders, not demographic parity in job recommendations. Please verify the citation or use a more directly relevant source.
  4. [Abstract] The abstract says 'certification-oriented assessment,' while Section 2.3.2 and Section 3.5 stress that the Fraunhofer Catalog is not formal certification. Consider 'assessment-oriented' to avoid a misleading implication.
  5. [3.2.1] The three inventory outcomes 'approved, requires correction, blocked pending review' could be defined more precisely; in particular, the difference between 'requires correction' and 'blocked pending review' is not clear from the description of hard constraints.

Circularity Check

0 steps flagged

No circular derivation: the framework is conceptual, assembled from external prior work; self-citations are supporting evidence, not load-bearing premises.

full rationale

The paper is a conceptual framework proposal, not a derivation with fitted parameters, predicted quantities, or equations whose outputs reduce to their inputs. Its central architectural claim — a two-stage hard/soft constraint handoff linking skill-extraction bias assessment to multistakeholder recommendation under an AI Act-oriented audit layer — is constructed from external building blocks: the Fraunhofer AI Assessment Catalog (external), social-choice aggregation (Aird et al., Brandt et al., Dignum et al., Popescu), and distributional auditing/counterfactual testing (Adhikari et al., Hort et al., Iso et al., Scher et al.). The self-citations ([4], [5], [6], [19], [33]) are used as supporting evidence that assessment catalogs have practical utility, that iterative assessment can surface technical problems, and that long-term fairness dynamics matter; they are not invoked as a uniqueness theorem, nor do they supply the framework's main conceptual move. The paper explicitly disclaims validation: in Section 4 it states 'The framework remains conceptual and has not been evaluated as a complete system' and 'The precise objectives, thresholds, and permissible interventions of the handoff remain design decisions requiring empirical validation.' The threshold gap a skeptical reader identifies is a missing operating-point procedure, not a definitional reduction: the paper never defines 'fairness' as 'whatever its own thresholds enforce,' so the proposed governance layer is not true by construction. Under the review rules, the manuscript's own limitation statements are weighed and they support a non-circularity verdict: the weakness is absence of empirical validation, not circular reasoning. No specific circular step can be quoted or exhibited, so the appropriate score is 0.

Axiom & Free-Parameter Ledger

2 free parameters · 5 axioms · 2 invented entities

No numerical parameters are fitted; the framework is conceptual. The central claim rests on several unvalidated domain assumptions about the reliability of auditing procedures and aggregation mechanisms, and on unspecified thresholds/weights. No new physical or mathematical entities are introduced; the Regulator agent and fairness-state ledger are software/regulatory constructs without independent evidence.

free parameters (2)
  • Fairness thresholds (Stage 1 and Stage 2) = not specified
    Hard/soft constraint classification depends on predefined thresholds for distributional skew and fairness metrics; no values or calibration procedure are given, and Section 4 calls them design decisions.
  • Agent weights / aggregation rule = not specified
    Social-choice aggregation results depend on voting rule, agent weights, and tie-breaking; the paper says the choice is use-case dependent but does not specify how to set them.
axioms (5)
  • domain assumption Historical labor-market data contain measurable protected-attribute distributions that reliably indicate bias (distributional auditing produces a valid ground truth).
    Section 3.1.2 proposes recording skill assignments by demographic group and flagging deviations; this presumes protected attributes are available/observable and that population parity is the right benchmark.
  • domain assumption Counterfactual gender-expression substitution and stereotype injection reveal model-internalized bias without introducing artifacts.
    Section 3.1.2's two counterfactual procedures are asserted to probe internalized associations; their validity for skill extraction is not empirically demonstrated.
  • domain assumption Social-choice aggregation of independent stakeholder rankings yields a recommendation that is fairer and more auditable than a single scoring function.
    Section 3.3.3 relies on social choice theory for transparent aggregation, but fairness of the aggregate is not guaranteed and Section 4 accepts this.
  • domain assumption The Fraunhofer AI Assessment Catalog is an appropriate tool for AI Act compliance preparation for this pipeline.
    Section 3.5 assumes the catalog's evidence requirements map onto the proposed tests; the paper acknowledges it is not formal conformity assessment.
  • domain assumption Chatbot-based ingestion is the representative high-risk input path; conclusions generalize to CV and manual entry.
    Section 3.1 focuses on chatbots and leaves other paths to future work, yet the framework claims system-wide coverage.
invented entities (2)
  • Regulator agent no independent evidence
    purpose: A stakeholder agent that re-ranks candidate-job matches according to fairness thresholds and tracks persistent fairness states for protected groups.
    No implementation or empirical demonstration; its behavior and thresholds are unspecified, and fairness objectives are inherited from Stage 1's soft constraints.
  • Persistent fairness states (bias reports ledger) no independent evidence
    purpose: A stored record of soft-constraint deviations across recommendation cycles to detect drift.
    A conceptual ledger; no data structure, semantics, or validation provided.

reviewed 2026-08-01 · how reviews work

0 comments
Cite this review

Pith. "Pith review of From Skill Extraction to Multistakeholder Recommendation: A Two-Stage Framework for Bias Governance in Skills-Based Job Matching." pith.science (2026). https://pith.science/paper/T5YOUCOU

@misc{pith2026260715707,
  author       = {Pith},
  title        = {Pith review of: From Skill Extraction to Multistakeholder Recommendation: A Two-Stage Framework for Bias Governance in Skills-Based Job Matching},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/T5YOUCOU}},
  note         = {Machine review of arXiv:2607.15707}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

AI-based labor-market systems or platforms can affect access to job opportunities prior to organizational candidate rankings or hiring decisions. Such applications warrant caution, as biases in skill extraction, profile formation, and candidate-job matching may contribute to unfair treatment of candidates. In this paper, we propose a two-stage framework for detecting and governing bias in skills-based job matching. Stage 1, skill extraction and profile formation, addresses how candidates provide skills and preferences to the system, how the system extracts and structures this information, and the bias risks this entails, with a focus on chatbot-based elicitation. Stage 2, multistakeholder candidate-job recommendation, would embed this information in a recommender system in which candidate, company, and regulatory objectives are represented by separate agents, each producing an independent candidate-job ranking; these rankings would be combined through social choice-based aggregation into a single, auditable recommendation. The two stages are connected by a shared distinction between hard constraints, which require correction before processing continues, and soft constraints, which are logged to inform later decisions. Following an AI Act-aligned assessment methodology (based on the Fraunhofer AI Assessment Catalog), we propose using distributional auditing and counterfactual testing to produce a Stage 1 bias inventory sorted into hard and soft constraints, with the latter informing fairness thresholds for Stage 2. The same logic would apply to Stage 2: fairness metrics crossing predefined thresholds would trigger an adapted recommendation process, while smaller deviations would be logged as bias reports and persistent fairness states.

Figures

Figures reproduced from arXiv: 2607.15707 by Andrea Forster, Dominik Kowald, Gregor Autischer, Simone Kopeinik.

Figure 1
Figure 1. Figure 1: Two-stage conceptual framework for bias detection and governance in [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

36 extracted references · 4 linked inside Pith

  1. [1]

    Abdollahpouri, H., Mansoury, M., Burke, R., and Mobasher, B. (2019). The impact of popularity bias on fairness and calibration in recommendation. arXiv preprint arXiv:1910.05755

    Pith/arXiv arXiv 2019

  2. [2]

    Adhikari, A., Vethman, S., Vos, D., Lenz, M., Cocu, I., Tolios, I., and Veenman, C. J. (2024). Gender mobility in the labor market with skills-based matching models.AI and Ethics, 4(1):163–167

    2024

  3. [3]

    Aird, A., Farastu, P., Sun, J., Stefancova, E., All, C., Voida, A., Mattei, N., and Burke, R. (2024). Dynamic fairness-aware recommendation through multi-agent social choice.ACM TORS, 3(2):1–35

    2024

  4. [4]

    Autischer, G., Waxnegger, K., and Kowald, D. (2025a). AI Certification and Assessment Catalogues: Practical Use and Challenges in the Context of the European AI Act. InProceedings of Fourth European Workshop on Algorithmic Fairness, pages 492–498. PMLR. ISSN: 2640-3498

  5. [5]

    Autischer, G., Waxnegger, K., and Kowald, D. (2025b). Practical applica- tion and limitations of AI certification catalogues in the light of the AI act. arXiv:2502.10398

    Pith/arXiv arXiv

  6. [6]

    Autischer, G., Waxnegger, K., and Kowald, D. (2026). Towards EU AI Act Compliance: Self-Certification and Fairness Alignment for Facial Emotion Recognition. InProceedings of Fifth European Conference on Algorithmic Fairness. PMLR. ECAF’26

    2026

  7. [7]

    Baum, K., Bryson, J., Dignum, F., Dignum, V., Grobelnik, M., Hoos, H., Irgens, M., Lukowicz, P., Muller, C., Rossi, F., Shawe-Taylor, J., Theodorou, A., and Vinuesa, R. (2023). From fear to action: AI governance and oppor- tunities for all.Frontiers in Computer Science, 5

    2023

  8. [8]

    Brandt, F., Conitzer, V., Endriss, U., Lang, J., and Procaccia, A. D. (2016). Handbook of computational social choice. Cambridge University Press

    2016

  9. [9]

    Burke, R. (2017). Multisided fairness for recommendation.Workshop on Fairness, Accountability, and Transparency in Machine Learning

    2017

  10. [10]

    Chatard, Y., Riede, L., Werkmeister, C., Ehlen, T., Roos, P., Kirchmair, V., and Knoke, L. (2024). EU AI act unpacked #10: ISO 42001 - a tool to achieve AI act compliance?

    2024

  11. [11]

    Cheong, B. C. (2024). Transparency and accountability in AI systems: safeguarding wellbeing in the age of algorithmic decision-making.Frontiers in Human Dynamics, 6

    2024

  12. [12]

    Non-discrimination

    Council of the European Union (2025). Non-discrimination. Consilium. Retrieved April 21, 2026, fromhttps://www.consilium.europa.eu/en/ policies/non-discrimination/. 16

    2025

  13. [13]

    Deldjoo, Y. (2025). Understanding biases in chatgpt-based recommender systems: Provider fairness, temporal stability, and recency.ACM TORS, 4(2):1–35

    2025

  14. [14]

    and Dignum, F

    Dignum, V. and Dignum, F. (2025). Agentifying agentic ai.arXiv preprint arXiv:2511.17332

    arXiv 2025

  15. [15]

    Escobedo, G., Moscati, M., Muellner, P., Kopeinik, S., Kowald, D., Lex, E., and Schedl, M. (2024). Making alice appear like bob: A probabilistic preference obfuscation method for implicit feedback recommendation models. InJoint European Conference on Machine Learning and Knowledge Discovery in Databases, pages 349–365. Springer

    2024

  16. [16]

    Key issue 2: Conformity assessment and self-assessment

    EU AI Act (2024). Key issue 2: Conformity assessment and self-assessment

    2024

  17. [17]

    ESCO (European Skills, Competences, Qualifications and Occupations) portal

    European Commission (2026). ESCO (European Skills, Competences, Qualifications and Occupations) portal. Accessed: 2026-07-08

    2026

  18. [18]

    Regulation (EU) 2024/1689 of the European Par- liament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act)

    European Union (2024). Regulation (EU) 2024/1689 of the European Par- liament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union

    2024

  19. [19]

    Forster, A., M¨ ullner, P., Helic, D., Lex, E., and Kowald, D. (2026). Fair agents: Balancing multistakeholder alignment in multi-agent personalization systems.arXiv preprint arXiv:2605.02379

    Pith/arXiv arXiv 2026

  20. [20]

    Ge, Y., Liu, S., Gao, R., Xian, Y., Li, Y., Zhao, X., Pei, C., Sun, F., Ge, J., Ou, W., et al. (2021). Towards long-term fairness in recommendation. InProceedings of the 14th ACM international conference on web search and data mining, pages 445–453

    2021

  21. [21]

    Conformity assessments in the EU AI act: What you need to know

    Holistic AI (2023). Conformity assessments in the EU AI act: What you need to know

    2023

  22. [22]

    M., Harman, M., and Sarro, F

    Hort, M., Chen, Z., Zhang, J. M., Harman, M., and Sarro, F. (2024). Bias mitigation for machine learning classifiers: A comprehensive survey.ACM Journal on Responsible Computing, 1(2):1–52

    2024

  23. [23]

    Iso, H., Pezeshkpour, P., Bhutani, N., and Hruschka, E. (2025). Evaluating bias in llms for job-resume matching: Gender, race, and education. InPro- ceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technolo- gies (Volume 3: Industry Track), pages 672–683

    2025

  24. [24]

    and Bogers, T

    Kaya, M. and Bogers, T. (2025). Mapping stakeholder needs to multi-sided fairness in candidate recommendation for algorithmic hiring. InProceedings of RecSys’25, pages 257–267. 17

    2025

  25. [25]

    Kilian, R., Jack, L., and Ebel, D. (2025). European AI standards: Tech- nical standardization and implementation challenges under the EU AI act. European Journal of Risk Regulation, 16(3):1038–1062

    2025

  26. [26]

    Lesota, O., Melchiorre, A., Rekabsaz, N., Brandl, S., Kowald, D., Lex, E., and Schedl, M. (2021). Analyzing item popularity bias of music recommender systems: are different genders equally affected? InProceedings of RecSys’21, pages 601–606

    2021

  27. [27]

    Mueck, M., Cadzow, S., and Wood, S. (2022). ETSI activities in the field of artificial intelligence: Preparing the implementation of the european AI act. Technical report, ETSI

    2022

  28. [28]

    Pimentel, B. (2024). Why AI still needs regulation despite impact

    2024

  29. [29]

    Popescu, G. (2013). Group recommender systems as a voting problem. InInternational Conference on Online Communities and Social Computing, pages 412–421. Springer

    2013

  30. [30]

    B., Hecker, D., Houben, S., Rosenzweig, J., Sicking, J., Schulz, E., Voss, A., and Wrobel, S

    Poretschkin, M., Schmitz, A., Akila, M., Adilova, L., Becker, D., Cremers, A. B., Hecker, D., Houben, S., Rosenzweig, J., Sicking, J., Schulz, E., Voss, A., and Wrobel, S. (2023). AI Assessment Catalog. Technical report, Fraunhofer IAIS

    2023

  31. [31]

    Rus, C., Mansoury, M., Yates, A., and de Rijke, M. (2026). Joint modeling of candidate and recruiter preferences for fair two-sided job matching. In European Conference on Information Retrieval, pages 335–351. Springer

    2026

  32. [32]

    Auditing machine learning algorithms

    SAI-FI-DE-NL-NO-UK (2023). Auditing machine learning algorithms. Technical report, Supreme Audit Institutions FI, DE, NL, NO, UK

    2023

  33. [33]

    Scher, S., Kopeinik, S., Tr¨ ugler, A., and Kowald, D. (2023). Modelling the long-term fairness dynamics of data-driven targeted help on job seekers. Scientific Reports, 13(1):1727

    2023

  34. [34]

    How ISO 42001 helps with EU AI act compliance

    Vanta (2025). How ISO 42001 helps with EU AI act compliance

    2025

  35. [35]

    Werry, S., Ridgway, S., Kerr-Shaw, S., and Silverstein, M. (2024). EU standardization supporting the artificial intelligence act

    2024

  36. [36]

    M., Eder, S., Weissenb¨ ock, J., Schwald, C., Doms, T., Vogt, T., Hochreiter, S., and Nessler, B

    Winter, P. M., Eder, S., Weissenb¨ ock, J., Schwald, C., Doms, T., Vogt, T., Hochreiter, S., and Nessler, B. (2021). Trusted Artificial Intelligence: Towards Certification of Machine Learning Applications.arXiv preprint arXiv:2103.16910. 18

    Pith/arXiv arXiv 2021

This paper was first reviewed by deepseek-v4-flash on August 1, 2026.