Possible harms of artificial intelligence and the EU AI act: fundamental rights and risk
Open accessThe DOI matches its Crossref record; the references it cites are not checked yet; none of the 38 sources its claims cite is checked for support yet.
Abstract
Various actors employ the notion of risk when they discuss the future role of Artificial Intelligence (AI) in society – sometimes as a general pointer to possible unwanted consequences of the underlying technologies, sometimes oriented towards a political regulation of AI risks. Mostly discussed within a legal or ethical framework, we still lack a perspective on AI risks based on sociological risk research. Building on systems-theoretical thinking about risk and society, this article analyses the potential and limits of a risk-based regulation of AI, in particular with regard to the notion of harm to fundamental rights. Drawing on the AI Act, its earlier drafts and related documents, the paper analyses how this regulatory framework delineates harms of AI and which implications the chosen delineation has for the regulation. The results show that fundamental rights are invoked as legal rules, as values and as a foundation for trustworthiness of AI in parallel to being identified as at risk from AI. The attempt to frame all possible harms in terms of fundamental rights creates communicative paradoxes. It opens the door to a political classification of high-risk AI systems as well as a future standard-setting that is removed from systematic concerns about fundamental rights and values. The additional notion of systemic risk, addressing possible risks from general-purpose AI models, further reveals the problems with delineating harms of AI. In sum, the AI Act is unlikely to achieve what it aims to do, namely the creation of conditions for trustworthy AI.
Overview
CoChat Summary
TL;DR
Building on Luhmannian systems-theoretical risk sociology, this article analyses how the EU AI Act delineates the harms of AI through the notion of risks to fundamental rights, and argues that this framing produces communicative paradoxes. The Act is unlikely to achieve its stated aim of creating conditions for trustworthy AI.
Key findings
- Fundamental rights are invoked in three parallel ways in the AI Act and related documents — as legal rules, as basic values, and as the foundation of trustworthiness — while simultaneously being identified as what AI puts at risk. Each parallel use generates a paradox: law is supposed to reduce uncertainty while the regulation concedes AI makes the future legally more uncertain; values are both the object of protection and the implicit justification for high-risk classifications; and the reason to trust AI (respect for fundamental rights) is undermined by the reason trust is needed (AI may harm those same rights).
- The high-risk category is not the output of technical risk assessment. Art. 6 and Annex III turn potential harm to fundamental rights into a discrete binary variable with no quantification and no stated rule from which the Annex III items are derived; Art. 7(1) permits additions where risk is "equivalent to or greater than" existing entries, with no method for determining equivalence. The category is therefore "the result of political judgments that are implicitly linked to values."
- Specification is delegated to the European Commission and eventually to standardisation bodies, which per Veale and Borgesius (2021) have no experience with fundamental rights, so value prioritisation is likely to occur implicitly and invisibly, and may be irreversible in practice once standards are set.
- The late addition of systemic risk for general-purpose AI models (Art. 52a; Rec. 60m–60o) adds an emphasis on scale but does not resolve the problem: each example ("disruptions of critical sectors," "foreseeable negative effects on democratic processes," "dissemination of illegal, false, or discriminatory content") requires further value judgements about what counts as serious or undesirable, and the Codes of Practice process is unlikely to address the ordering of values systematically.
Methods
- Qualitative document analysis using the dimensions of social meaning from Luhmann's theory of communication (time, factual, social) as overarching categories for a close reading of how a delineation of AI harms is attempted.
- Key documents: the European Commission's original draft AI Act of 21 April 2021 including annexes, the Council of the EU revision of 6 December 2022, the European Parliament's amended draft of 14 June 2023, and a leaked four-column trilogue comparison document made public in January 2024; plus the 2020 White Paper on AI and the 2019 Ethics Guidelines for Trustworthy AI, referenced by the draft's explanatory memorandum.
- The author states the goal was not legal but sociological analysis, so interest was in implied basic understandings of risk rather than specific wordings.
Limitations
- Stated by the paper: it does not offer a comprehensive account of the AI Act and its legal and political implications; the perspective is selective, focusing on the legislation's content from the standpoint of sociological risk research; and, written after agreement on the text but before it entered into force, the analysis is limited to what the AI Act, its earlier drafts and key reference documents communicate. The author calls for future research on the Codes of Conduct and standards.
- Evident from the text: the analysis is conceptual and interpretive rather than empirical, with no data, sample or quantitative results; it relies in part on a leaked trilogue document rather than an official published text; and its central claims about future standard-setting and value prioritisation are projections rather than observed outcomes.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
0 Valid · 0 Not valid · 34 Not checked · 78 Own results and interpretations
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The results show that fundamental rights are invoked as legal rules, as values and as a foundation for trustworthiness of AI in parallel to being identified as at risk from AI.”
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
In the paper
“The amendments of the European Parliament (Citation2023) added democracy, the rule of law and the environment to the list of entities potentially at risk from AI.”
What the sources say
- Not checked yetcites European Parliament Citation2023
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In sum, the AI Act is unlikely to achieve what it aims to do, namely the creation of conditions for trustworthy AI.”
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The attempt to frame all possible harms in terms of fundamental rights creates communicative paradoxes.”
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
None of the sources cited for this claim have been checked yet.
In the paper
“There is no objective way to assess and measure risks.”
What the sources say
- Not checked yetcites Boholm and Corvellec Citation2011
- Not checked yetcites Christoffersen Citation2018
- Not checked yetcites Hilgartner Citation1992
plain- Asserted: passes — The authors put it forward as true.
- Checkable: unclear — It is unclear whether it could be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
19 more claims
Create a free account to see all 112 claims and to verify the 38 references still open.
- Probably not a substantial claim. 78
- It may state two separate claims as one. 33
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?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Kusche identifies the danger that fundamental rights may lose part of their normative force when they are translated into risks to be balanced and managed rather than treated as binding limits on public power (Kusche, 2024).”
From Human-centric AI governance in the European Union. Accountability, fundamental rights, and institutional resilience in public administration (2026) · cites this paper as Kusche, 2024
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“Work on trustworthy AI, standards, and risk informed the distinction between formal compliance and institutional capability(European Commission, 2019;Gamito & Marsden, 2024;Kusche, 2024;Laux et al., 2024).”
From Responsible AI in Polish Higher-Education Administration under the EU AI Act: High-Risk Use Cases, Fundamental-Rights Impact Assessment, and Institutional Accountability (2026) · cites this paper as Kusche, 2024
Citations
The citation lists are queued
They are fetched later and appear here then.
References 45 · Cited by 87
Research works printed but not linked: 7
- ‘CDT Europe’s AI Bulletin: March 2023’. 2023. Center for Democracy and Technology (blog). March 12 2023. https://cdt.org/insights/cdt-europes-ai-bulletin-march-2023/.
- Joint Research Centre. 2023. “Analysis of the Preliminary AI Standardisation Work Plan in Support of the AI Act.” LU: European Commission. https://data.europa.eu/doi/10.2760/5847. · doi:10.2760/5847
- Kaminski Margot E. 2023. “Regulating the Risks of AI.” Boston University Law Review 103 (5): 1347–1411.
- Luhmann, Niklas. 2005. Risk: A Sociological Theory. 1st paperback ed. New Brunswick, N.J: Aldine Transaction.
- Luhmann, Niklas. 2017. Trust and Power. English ed. Malden, MA: Polity.
- 10.6094/behemoth.2009.2.3.705 · doi:10.6094/behemoth.2009.2.3.705
- 10.1023/A:1021414628203 · doi:10.1023/a:1021414628203
Metadata
- Authors
- Kusche, I.
- Affiliations
- University of Bamberg (DE)
- Corresponding author
- Kusche, Isabel
- ORCID
- Kusche, Isabel
- Venue
- Journal of Risk Research
- Publisher
- Informa UK Limited
- Published
- 2024-05-11
- Pages
- 1-14
- ISSN
- 1366-9877 (ISSN-L)
- 1466-4461
- OpenAlex
- W4396826430
- Cited by
- 87
- References
- 45
- Citations by year
- 2026: 53 · 2025: 32 · 2024: 2
- FWCI
- 224.93
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- Locations
- Journal of Risk Research (publishedVersion, cc-by-nc-nd)
- Topics
- Social Sciences › Law › Digitalization, Law, and Regulation
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Social Sciences › Business, Management and Accounting › Industrial relations › Legal and Labor Studies
- Keywords
- artificial intelligence, EU AI Act, AI harms, fundamental rights, risk-based regulation, systems theory, trustworthy AI, high-risk AI systems, systemic risk, general-purpose AI models, AI regulation, risk classification
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.
A language model drew the claims from the paper on 2026-10-05, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-05 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Possible harms of artificial intelligence and the EU AI act: fundamental rights and risk. https://app.cochat.ai/papers/10.1080/13669877.2024.2350720
- The paper
- Kusche, I. (2024). Possible harms of artificial intelligence and the EU AI act: fundamental rights and risk. Journal of Risk Research. https://doi.org/10.1080/13669877.2024.2350720
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.