---
title: "Legal Considerations in the Use of Synthetic Data for AI"
description: "This position paper by ELDA and ALT-EDIC authors asks when synthetic data can lawfully support the fine-tuning of… 10 references."
authors: ["Kossay Talmoudi", "Khalid Choukri", "Amélie Gourgeot", "Florine Astruc"]
year: 2026
venue: "Proceedings of the Joint Workshop on Legal and Ethical Issues in Human Language Technologies and Computational Approaches to Language Data Pseudonymization, Anonymization, De-identification, and Data Privacy (LEGAL2026 and CALD-pseudo 2026) @ LREC 2026"
source: "https://aclanthology.org/2026.legal-1.10/"
doi: "10.63317/3vwisn8odtmp"
licence: "CC BY-NC 4.0 (ELRA, LREC proceedings)"
licence_url: "https://creativecommons.org/licenses/by-nc/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/talmoudi-2026-legal-considerations-use-synthetic-data/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/talmoudi-2026-legal-considerations-use-synthetic-data/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/34/) / Legal Considerations in the Use of Synthetic Data for AI Development and Finetuning: The Case of LLMs4EU

# Legal Considerations in the Use of Synthetic Data for AI Development and Finetuning: The Case of LLMs4EU

**Authors:** Kossay Talmoudi, [Khalid Choukri](https://md.research.elamur.ai/free/en/authors/khalid-choukri/), Amélie Gourgeot and Florine Astruc · **Year:** 2026 · **Venue:** [Proceedings of the Joint Workshop on Legal and Ethical Issues in Human Language Technologies and Computational Approaches to Language Data Pseudonymization, Anonymization, De-identification, and Data Privacy (LEGAL2026 and CALD-pseudo 2026) @ LREC 2026](https://md.research.elamur.ai/free/en/venues/legal/) ([Joint Workshop on Legal and Ethical Issues in Human Language Technologies and Computational Approaches to Language Data Pseudonymization, Anonymization, De-identification, and Data Privacy @ LREC 2026](https://md.research.elamur.ai/free/en/venues/legal/2026/)) · **Publisher:** [ELRA](https://md.research.elamur.ai/free/en/publishers/elra/) · **Licence:** [CC BY-NC 4.0 (ELRA, LREC proceedings)](https://creativecommons.org/licenses/by-nc/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.legal-1.10/) · [PDF](https://aclanthology.org/2026.legal-1.10.pdf) · [DOI 10.63317/3vwisn8odtmp](https://doi.org/10.63317/3vwisn8odtmp) · [OpenAlex](https://openalex.org/W7166516342)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
This position paper by ELDA and ALT-EDIC authors asks when synthetic data can lawfully support the fine-tuning of general-purpose AI models in the EU. It uses the EU-funded LLMs4EU project as its case study. The authors frame synthetic text as an answer to the shrinking supply of human-written data, in line with EU data policy. They then work through three legal regimes.

- **GDPR:** they ask whether synthetic records count as personal data.
- **Copyright:** they assess the risk of memorised or infringing content passing from generator models into new corpora, citing the GEMA v. OpenAI ruling.
- **Contracts:** they examine how model licences and acceptable-use policies may forbid using one model's outputs to train another.

The paper runs no experiments; it reviews legal sources and empirical studies. Its conclusion is that synthetic data reduces legal risk but does not remove it. Liability shifts upstream to the generation process. Safe use therefore needs governance across the whole data chain: careful choice of generator models, licence checks, privacy and copyright audits, and documented generation. This matters for any team, multilingual NLP groups included, that relies on LLM-generated training data.

## Key points
- The paper cites projections (Villalobos et al., 2024) that public human-generated text could run out for large-scale LLM training by around 2032. It places synthetic data within the 2020 European Strategy for Data, the 2025 Data Union Strategy and the Language Data Space.
- Synthetic data is not defined in the Data Act, but its broad, technology-neutral notion of "data" in Article 2(1) covers it. Several legal regimes may therefore apply at once.
- On privacy, partially synthetic data remains exposed to membership inference, while fully synthetic data resists attacks better (Zhang et al., 2022). A 2026 mental-health study using three attack-based evaluations found low but non-zero linkage risk, which is enough for the GDPR to apply.
- The paper draws on EDPB Opinion 28/2024 on AI models and EDPB Guidelines 01/2025 on pseudonymisation. These support treating synthetic data as a privacy-enhancing tool, not a loophole around data protection rules.
- LLMs4EU is funded by the Digital Europe Programme (grant 10,119,847). It will treat synthetic data as one component of a governance framework that is periodically reassessed against new case law and technical evidence.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

📚 Shelves (5): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🤝 Social Sciences](https://md.research.elamur.ai/free/en/shelves/social-sciences/) · 1,772 works · [🤝 Safety Research](https://md.research.elamur.ai/free/en/shelves/social-sciences/safety-research/) · 543 works · [⚖️ Law](https://md.research.elamur.ai/free/en/shelves/social-sciences/law/) · 77 works

🧭 Topics (1): [⚖️ Legal AI](https://md.research.elamur.ai/free/en/topics/legal-ai/) · 134 works

## Cite

Cite as: Kossay Talmoudi, Khalid Choukri, Amélie Gourgeot and Florine Astruc. 2026. Legal Considerations in the Use of Synthetic Data for AI Development and Finetuning: The Case of LLMs4EU. Proceedings of the Joint Workshop on Legal and Ethical Issues in Human Language Technologies and Computational Approaches to Language Data Pseudonymization, Anonymization, De-identification, and Data Privacy (LEGAL2026 and CALD-pseudo 2026) @ LREC 2026.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{talmoudi2026legal,
  title = {{Legal Considerations in the Use of Synthetic Data for AI Development and Finetuning: The Case of LLMs4EU}},
  author = {Kossay Talmoudi, Khalid Choukri, Amélie Gourgeot and Florine Astruc},
  booktitle = {Proceedings of the Joint Workshop on Legal and Ethical Issues in Human Language Technologies and Computational Approaches to Language Data Pseudonymization, Anonymization, De-identification, and Data Privacy (LEGAL2026 and CALD-pseudo 2026) @ LREC 2026},
  year = {2026},
  doi = {10.63317/3vwisn8odtmp},
  url = {https://doi.org/10.63317/3vwisn8odtmp}
}
```

- [← previous LEDGER: Scaling Agentic Document Editing with Dependency-aware Graph Retrieval](https://md.research.elamur.ai/free/en/papers/wang-2026-ledger-scaling-agentic-document-editing/)
- [next → LegalChainReasoner: Grounding Criminal Judicial Opinion Generation via Structured Legal Chains](https://md.research.elamur.ai/free/en/papers/shi-2026-legalchainreasoner-grounding-criminal-judicial-opinion/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/34/)

## References

10 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Achterberg, J., van Dijk, B., Waseem, H.M., Gallos, P., Epiphaniou, G., Maple, C., Haas, M. Spruit, M. (2025). The Data Sharing Paradox of Synthetic Data in Healthcare. arXiv preprint arXiv:2503.20847. [↗](https://arxiv.org/abs/2503.20847)
2. Ahmed, A., Cooper, A. F., Koyejo, S., & Liang, P. (2026). Extracting books from production language models. arXiv preprint arXiv:2601.02671. [↗](https://arxiv.org/abs/2601.02671)
3. Carlini, N., Tramer, F., Wallace, E., Jagielski, M., Herbert-Voss, A., Lee, K., Roberts, A., Brown, T., Song, D., Erlingsson, U., Oprea, A., Raffel, C. (2021). [Extracting training data from large language models](https://md.research.elamur.ai/free/en/papers/carlini-2020-extracting-training-data-large-language/). In 30th USENIX security symposium (USENIX Security 21), pages 2633-2650. [↗](https://doi.org/10.48550/arxiv.2012.07805)
4. European Commission, Joint Research Centre, (2024) Synthetic data in the Data Hub of the Digital Finance Platform, Publications Office of the European Union, Luxembourg European Data Protection Board, (2025) Guidelines 01/2025 on Pseudonymisation European Data Protection Board, (2024) Opinion 28/2024 on certain data protection aspects of AI models Gal, M. Lynskey, O. (2023), Synthetic Data: Legal Implications of the Data-Generation Revolution, 109 Iowa Law Review Goodfellow, J. Pouget-Abadie, J. Mirza, M. Xu, B. Warde-Farley, D. Ozair, S. Courville, A. Bengio, Y. (2014) Generative adversarial networks, In Proceedings of the 27th International Conference on Neural Information Processing Systems, Volume 2, pages 2672-2680 [find](https://scholar.google.com/scholar?q=European+Commission%2C+Joint+Research+Centre%2C+%282024%29+Synthetic+data+in+the+Data+Hub+of+the+Digital+Finance+Platform%2C+Publications+Office+of+the+European+Union%2C+Luxembourg+European+Data+Protection+Board%2C)
5. Haizoune, M., Leventhal, B. L., Pant, D., Nytrø, Ø., Koochakpour, K., Koposov, R.A., Øhlckers, L.R., Skokauskas, N. (2026). Balancing Privacy and Utility in Child and Adolescent Mental Health Services Research: Retrospective Cohort Study on Synthetic Data Generation. JMIR Medical Informatics, 14, e71819. [↗](https://doi.org/10.2196/71819)
6. Rosati, E. (2025) Infringing AI: Liability for AI-generated outputs under international, EU, and UK copyright law, European Journal of Risk Regulation 16(2) p. 611 [find](https://scholar.google.com/scholar?q=Rosati%2C+E%2E+%282025%29+Infringing+AI%3A+Liability+for+AI%2Dgenerated+outputs+under+international%2C+EU%2C+and+UK+copyright+law%2C+European+Journal+of+Risk+Regulation+16%282%29+p%2E+611)
7. Shumailov, I., Shumaylov, Z., Zhao, Y., Papernot, N., Anderson, R., Gal, Y. (2024) AI models collapse when trained on recursively generated data. Nature 631, pages 755–759. [↗](https://doi.org/10.1038/s41586-024-07566-y)
8. Tyagi, K. (2025) Synthetic Data, Data Protection and Copyright in an era of Generative AI, 16 JIPITEC p.176 [find](https://scholar.google.com/scholar?q=Tyagi%2C+K%2E+%282025%29+Synthetic+Data%2C+Data+Protection+and+Copyright+in+an+era+of+Generative+AI%2C+16+JIPITEC+p%2E176)
9. Villalobos, P. Ho, A. Sevilla, J. Besiroglu, T. Heim, L. Hobbhahn, M. (2024) Will we run out of data? Limits of LLM scaling based on human-generated data, arXiv preprint arXiv:2211.04325 [↗](https://arxiv.org/abs/2211.04325)
10. Zhang, Z., Yan, C., & Malin, B. A. (2022). Membership inference attacks against synthetic health data. Journal of biomedical informatics, 125, 103977. [↗](https://doi.org/10.1016/j.jbi.2021.103977)

*Updated 1 November 2026*