The DOI matches its Crossref record; 24 of the 24 references it cites were found in a registry or an index; 7 of the 188 sources its claims cite were checked for support.
Generative AI and deepfakes: a human rights approach to tackling harmful content
Open accessAbstract
© 2024 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/ licenses/by/4.0/).
Overview
CoChat Summary
TL;DR
The paper argues that the EU AI Act’s deepfake provisions, if implemented without safeguards, could infringe privacy and free expression under Articles 8 and 10 ECHR and the GDPR. It proposes two amendments: mandate structured synthetic data for deepfake detection and classify AI intended for malicious deepfakes as “high-risk.”
Key findings
- The AIA’s deepfake rules (Article 50 and Recitals 132–137) impose transparency and labelling duties but lack clear definitions and safeguards; the paper argues this creates risks under the ECHR’s accessibility, foreseeability, rule of law, necessity, and proportionality principles, and under the GDPR.
- Malicious deepfake harms are reported as growing: a 550% increase in AI-manipulated photos between 2019 and 2023; identified deepfake adult videos rose 550% in 2023 to 95,820, with 303,640,207 views on seven of the top ten adult websites; non-consensual deepfakes surged 464%; 99% of this content features women while 48% of men have encountered deepfakes.
- Detection tools are cited with claimed performance: Sensity claims 98.8% accuracy for AI-generated faces/bodies, and Intel’s FakeCatcher claims 96% precision/accuracy in real time. The paper also notes false-positive and overfitting concerns, including monitoring systems failing to recognise newly created child abuse imagery.
- The paper proposes two key amendments: (1) mandate structured synthetic data for deepfake detection, and (2) classify AI intended for deepfake electoral disinformation, extortion, and AI sexual abuse content as “high-risk” under Article 7(2), with clear definitions, oversight, and safeguards.
Methods
- Doctrinal legal analysis of EU AI Act provisions (including Article 3(60), Article 50, Recitals 132–137, and risk classification), GDPR provisions (Articles 4, 6, 9, 22), ECHR Articles 8 and 10, and relevant ECtHR and CJEU case law.
- Uses case examples and secondary sources—such as *Clarkson v OpenAI*, *Glukhin v Russia*, *Poland v Parliament and Council*, and *Meta v Bundeskartellamt*—to assess compatibility and justify proposed amendments.
- No empirical data collection or quantitative modelling; reported numbers are drawn from external reports and studies.
Limitations
- The paper states that a $25 million deepfake fraud case falls outside its scope, that synthetic data is “not a silver bullet,” and that safety notices require further research.
- Evident from the text: the analysis is doctrinal and does not empirically test the proposed amendments or their implementation effects; it relies on secondary reports, news sources, and ongoing litigation.
- Evident from the text: the analysis is focused on EU/ECHR/GDPR frameworks, with limited comparative or global coverage.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
5 Valid · 14 Not valid · 163 Not checked · 87 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Deepfakes possess a diverse range of applications beyond malicious intent. They are employed in various industries, including visual e ff ects, entertainment, and education. For instance, deepfakes can be used to create realistic special e ff ects in movies, personalise digital avatars, or generate synthetic voices for individuals who have lost theirs (Forbes Magazine, July 22, 2019). Additionally”
What the sources say
- Not matchedcites Forbes Magazine, July 22, 2019
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Partly supportedcites Mirsky and Lee 2020, 1 sourceExperimental — accuracy not yet measured
The Introduction (p. 1) describes positive uses in education, historical-figure reanimation, and entertainment. Section 2.1 (p. 3) adds movie and video-game production and photograph improvement, supporting most of the listed applications. The passages do not establish virtual reality as an application. (LLM, conf=0.86)
plain- 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.
- Not matched
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“Additionally, the viral nature of deepfakes, easily disseminated and replicated across platforms and jurisdictions, renders removal e ff orts challenging and potentially ine ff ective (Brown et al. 2022).”
What the sources say
- Could not verifycites Brown et al. 2022 sourceExperimental — accuracy not yet measured
Our copy discusses the difficulty of removing sensitive information from text datasets (Section 5.1.1) and how language models can retain earlier versions of deleted online text (Section 6). Neither passage addresses deepfakes, their spread across platforms or jurisdictions, or the effectiveness of removing them. Searches for deepfakes and removal of rapidly shared synthetic media found no passage supporting the attributed claim. (LLM, conf=0.75)
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.
- Could not verify
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“However, evaluating a system ' s CO2e emissions through traditional environmental impact assessments poses challenges. Quantifying all necessary information -particularly regarding hardware, datacenters, energy mix, and subsequent data disclosure -can be di ffi cult due to limited availability or accessibility (Patterson et al. 2021).”
What the sources say
- Partly supportedcites Patterson et al. 2021 sourceExperimental — accuracy not yet measured
Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.
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.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Leveraging technologies such as blockchain (Hasan and Salah 2019), or privacy-enhancing digital fi ngerprints can transparently trace content origins (Barrington et al. 2023).”
What the sources say
- Supportedcites Hasan and Salah 2019 source
The Introduction describes an Ethereum blockchain-based solution that provides publicly accessible provenance and traceability of digital content to its publishing source. “Tracing a Video to Its Origin” adds that the provenance data is openly accessible through the Ethereum ledger. The separate claim about privacy-enhancing digital fingerprints is attributed to Barrington et al., not to Hasan and Salah. (LLM, conf=0.94)
- Partly supportedcites Barrington et al. 2023 sourceExperimental — accuracy not yet measured
The Introduction (Section I, p. 1) states that active detection techniques 'at the point of synthesis, embed a perceptible or imperceptible watermark into [22], or extract a perceptual fingerprint [22] from, synthetically-generated content' and that 'these watermarks/fingerprints can then be used to identify content once it is released into the wild'; the Related Work (Section I-A, p. 2) adds that recent efforts target 'fingerprints that can identify specific synthesis architectures [26]', and the Discussion (Section V) urges synthesizers to embed imperceptible watermarks (citing Adobe's Content Authenticity Initiative). This supports the provenance-tracing half of the claim, but nothing in our copy describes these fingerprints as 'privacy-enhancing' — the paper frames them purely as detection/provenance mechanisms (and even concedes watermarking 'is not a panacea'), so that qualifier is not attributable to this source. (LLM, conf=0.70)
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.
- Supported
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 ease with which manipulations can be created, particularly using generative AI tools, poses an existential threat to the integrity of information and trust in public discourse.”
plainstates a cause- 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.
103 more claims
Create a free account to see all 269 claims and every source verdict.
- Probably not a substantial claim. 99
- It may state two separate claims as one. 100
Citations
Cites
The Creation and Detection of Deepfakes: A Survey
Mirsky, Y., & Lee, W. (2020) · arXiv · doi:10.48550/arxiv.2004.11138 · Cited by 863
The Creation and Detection of Deepfakes: A Survey
Mirsky, Y., & Lee, W. (2022) · ACM Computing Surveys · doi:10.1145/3425780 · Cited by 863
Veale, M., & Zuiderveen Borgesius, F. (2021) · Computer Law Review International · doi:10.9785/cri-2021-220402 · Cited by 805
Carbon Emissions and Large Neural Network Training
Patterson, D., Gonzalez, J., Le, Q., Liang, C., Munguia, L.-M., Rothchild, D., So, D., Texier, M., & Dean, J. (2021) · arXiv · doi:10.48550/arxiv.2104.10350 · Cited by 486
Combating Deepfake Videos Using Blockchain and Smart Contracts
Hasan, H. R., & Salah, K. (2019) · IEEE Access · doi:10.1109/access.2019.2905689 · Cited by 339 · cited 1×
“Leveraging technologies such as blockchain (Hasan and Salah 2019), or privacy-enhancing digital fi ngerprints can transparently trace content origins (Barrington et al. 2023).”
What Does it Mean for a Language Model to Preserve Privacy?
Brown, H., Lee, K., Mireshghallah, F., Shokri, R., & Tramèr, F. (2022) · 2022 ACM Conference on Fairness Accountability and Transparency · doi:10.1145/3531146.3534642 · Cited by 165
Generative AI in EU Law: Liability, Privacy, Intellectual Property, and Cybersecurity
(2024) · doi:10.2139/ssrn.4694565 · Cited by 124
Deep Learning for Deepfakes Creation and Detection: A SurveyPreprint
Nguyen, T. T., Nguyen, M. C., Nguyen, D. T., Nguyen, D. T., & Nahavandi, S. (2020) · arXiv (Cornell University) · doi:10.6084/m9.figshare.12731039.v1 · Cited by 124
Generative AI in EU law: Liability, privacy, intellectual property, and cybersecurity
Novelli, C., Casolari, F., Hacker, P., Spedicato, G., & Floridi, L. (2024) · Computer Law & Security Review · doi:10.1016/j.clsr.2024.106066 · Cited by 124
Deepfakes: regulatory challenges for the synthetic society
van der Sloot, B., & Wagensveld, Y. (2022) · Computer Law & Security Review · doi:10.1016/j.clsr.2022.105716 · Cited by 114
Research works printed but not linked: 2
- Shattock, E. 2022. “Fake News in Strasbourg: Electoral Disinformation and Freedom of Expression in the European Court of Human Rights (ECtHR).” European Journal of Law and Technology. https://ejlt.org/index.php/ejlt/article/view/882.
- Thi Nguyena T. Quoc Viet Hung Nguyenb Dung Tien Nguyena Duc Thanh Nguyena Thien Huynh-Thec Saeid Nahavandid Thanh TamNguyene Quoc-Viet Phamf and Cuong M. Nguyeng. 2022. “Deeplearning for Deepfakes Creation and Detection: A Survey.” https://www.sciencedirect.com/science/article/abs/pii/S1077314222001114.
Cited by
The Integration of AI and Metaverse in Education: A Systematic Literature Review
Almeman, K., Ayeb, F. E., Berrima, M., Issaoui, B., & Morsy, H. (2025) · Applied Sciences · doi:10.3390/app15020863 · Cited by 119
Park, S., & Nan, X. (2026) · AI & SOCIETY · doi:10.1007/s00146-025-02620-3 · Cited by 58
“The ease of creating convincing deep-fakes, removing the barrier of technical expertise, further amplifies the threat (Romero Moreno 2024).”
Liu, M., Zhang, L. J., & Zhang, D. (2025) · Computers in Human Behavior · doi:10.1016/j.chb.2025.108569 · Cited by 43
From social media to artificial intelligence: improving research on digital harms in youth
Mansfield, K. L., Ghai, S., Hakman, T., Ballou, N., Vuorre, M., & Przybylski, A. K. (2025) · The Lancet Child & Adolescent Health · doi:10.1016/s2352-4642(24)00332-8 · Cited by 55
Future Shock: Generative AI and the International AI Policy and Governance Crisis
Leslie, D., & Perini, A. M. (2024) · Harvard Data Science Review · doi:10.1162/99608f92.88b4cc98 · Cited by 33
Metadata
- Authors
- Romero Moreno, F.
- Affiliations
- Hertfordshire Schools of Law and Education, University of Hertfordshire, Hatfield, UK
- Venue
- International Review of Law, Computers & Technology
- Publisher
- Informa UK Limited
- Published
- 2024-03-29
- Volume
- 38
- Issue
- 3
- Pages
- 297-326
- ISSN
- 1360-0869 (ISSN-L)
- 1364-6885
- OpenAlex
- W4393308240
- Cited by
- 117
- References
- 20
- Citations by year
- 2026: 47 · 2025: 48 · 2024: 22
- FWCI
- 298.26
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- public-domain
- creativecommons.org/licenses/by/4.0/ (vor)
- Locations
- International Review of Law Computers & Technology (publishedVersion, public-domain)
- University of Hertfordshire Research Archive (University of Hertfordshire) (submittedVersion, cc-by)
- University of Hertfordshire Research Archive (University of Hertfordshire) (submittedVersion)
- Topics
- Social Sciences › Law › Digitalization, Law, and Regulation
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Physical Sciences › Computer Science › Artificial Intelligence › Law, AI, and Intellectual Property
- Keywords
- deepfakes, generative AI, EU AI Act, AI regulation, human rights, privacy, freedom of expression, General Data Protection Regulation, synthetic data, deepfake detection, high-risk AI classification, voter manipulation, extortion, misinformation, child online safety, democratic integrity
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.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-12, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-12 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol and qwen3.8-omni-flash).
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: Generative AI and deepfakes: a human rights approach to tackling harmful content. https://app.cochat.ai/papers/10.1080/13600869.2024.2324540
- The paper
- Romero Moreno, F. (2024). Generative AI and deepfakes: a human rights approach to tackling harmful content. International Review of Law, Computers & Technology. https://doi.org/10.1080/13600869.2024.2324540
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.


