Predictability of algorithmic behavior of artificial intelligence as a criterion of civil liability and regulatory control

Objective: to substantiate the conceptual limitations of the risk-based approach applied to civil liability for harm caused by artificial intelligence, and to provide arguments for the criterion of algorithmic behavior predictability.

Methods: based on the fiction of the complete controllability of algorithmic systems shaped within the legal doctrine, the author substantiates a comprehensive methodology combining comparative legal studies (comparing the risk-based AI Act model with Russian law), formal legal analysis (Articles 1, 10, 15, 393, 401, 404, 1079, 1083 of the Russian Civil Code) and hermeneutical method for transferring technical concepts (“hallucinations”, “reproducibility”) into legally relevant categories.

Results: the authors raises the question of the difference between risk and predictability: risk describes the probability and scale of harm (post hoc), while predictability characterizes the controllability of the system behavior ex ante. The authors substantiates that predictability allows one to transform engineering parameters (autonomy, stochasticity) into legally relevant criteria for assessing integrity and reasonableness (Articles 1, 401 of the Russian Civil Code). Using the example of “hallucinations” of generative models, the authors demonstrates that statistically expected deviations qualify within the framework of ordinary business risk, while going beyond the documented limits of testing may indicate a violation of the requirements for system manageability. The research shows that low predictability with high autonomy undermines the grounds of subjective liability and justifies the “source of increased danger” regime (Art. 1079 of the Russian Civil Code): differentiation of liability corresponds to the graduated nature of predictability (Articles 404, 1083 of the Russian Civil Code).

Scientific novelty: the authors substantiates the need to shift the focus from retrospective distribution of harm to a preliminary assessment of decisions on the algorithmic systems implementation, by institutionalizing algorithmic due diligence procedures, including documented evidence of predictability.

Practical significance: an algorithm for judicial assessment of three groups of circumstances is proposed (information about the system, empirical verification, organizational control mechanisms), as well as a recommendation to stipulate in a dedicated AI law the requirement to ensure and assess the predictability of high-risk AI systems.

1. Amodei, D., Olah, C., Steinhardt, J., Christiano, P., Schulman, J., & Mané, D. (2016). Concrete Problems in AI Safety. arXiv:1606.06565v2 [cs.AI]. https://arxiv.org/pdf/1606.06565

2. Baldwin, R., Cave, M., & Lodge, M. (2011). Risk-based Regulation. In Understanding Regulation: Theory, Strategy, and Practice (2nd ed. Chapter 13, pp. 281–295). Oxford: Oxford University Press. https://doi.org/10.1093/acprof:osobl/9780199576081.003.0013

3. Benk, M., et al. (2022). The value of measuring trust in AI – A socio-technical system perspective, 12. arXiv:2204.13480. https://doi.org/10.48550/arXiv.2204.13480

4. Bostrom, N. (2014). Superintelligence: Paths, Dangers, Strategies. Oxford: Oxford University Press. Brownsword, R. (2025). Law, Technology and Governance (1st ed.) Routledge. https://doi.org/10.4324/9781003658917

5. Dignum, V. (2019). Responsible Artificial Intelligence: How to Develop and Use AI in a Responsible Way. Springer. https://doi.org/10.1007/978-3-030-30371-6

6. Erakhtina, O. S. (2022). Risk-based approach to legal regulation of relations in the sphere of development and application of artificial intelligence technologies. Zakony Rossii: Opyt, Analiz, Praktika, 10, 101–104. (In Russ.).

7. Fraser, H., & Bello y Villarino, J.-M. (2023). Acceptable risks in Europe’s proposed AI Act: Reasonableness and other principles for deciding how much risk management is enough. arXiv:2308.02047. https://doi.org/10.48550/arXiv.2308.02047

8. Gawlikowski, J., Njieutcheu, C. R., Tassi, F., Ali, M., Lee, J., Humt, M., Feng, J., Kruspe, A., Triebel, R., Jung, P., Roscher, R., Shahzad, M., Yang, W., Bamler, R., & Zhu, X. X. (2023). A survey of uncertainty in deep neural networks. Neural Computing and Applications, 56, 1513–1589. https://doi.org/10.1007/s10462-023-10562-9

9. Gellert, R. (2020). Fundamental notions: Risk and regulation. In: The Risk-Based Approach to Data Protection. Oxford: Oxford University Press (online). https://academic.oup.com/book/40487/chapter-abstract/347671254

10. Goldberg, Y., Jacovi, A., Marasović, А., & Miller, Т. (2021). Formalizing trust in artificial intelligence: Prerequisites, causes and goals of human trust in AI. In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency. https://doi.org/10.1145/3442188.3445923

11. Grabosky, P. (2017). Meta-regulation. In P. Drahos (Ed.), Regulatory theory: Foundations and applications (pp. 149–171). Canberra: ANU Press.

12. Grieman, K., & Early, J. (2023). A Risk-based Approach to AI Regulation: System Categorisation and Explainable AI Practices. SCRIPTed, 20(1), 56–88.

13. Hildebrandt, M. (2018, September 13). Algorithmic regulation and the rule of law. Philosophical Transactions A: Mathematical, Physical and Engineering Sciences, 376(2128). https://doi.org/10.1098/rsta.2017.0355

14. Hüllermeier, E., & Waegeman, W. (2019). Aleatoric and Epistemic Uncertainty in Machine Learning: An Introduction to Concepts and Methods. arXiv:1910.09457. https://arxiv.org/abs/1910.09457

15. Karapetov, A. G. (Ed.). (2020). Main provisions of civil law: an article-by-article commentary to Articles 1-16.1 of the Russian Civil Code. Moscow: M-Logos. (In Russ.).

16. Königs, P. (2022) Artificial intelligence and responsibility gaps: what is the problem? Ethics and Information Technology, 24(36), 11. https://doi.org/10.1007/s10676-022-09643-0

17. Kustitskaya, T. A., Esin, R. V., & Noskov, M. V. (2025). Model Drift in Deployed Machine Learning Models for Predicting Learning Success. Computers, 14(9), 351. https://doi.org/10.3390/computers14090351

18. Kuteynikov, D., & Izhaev, O. (2023). Analysing Risk-Based Approach in the Draft EU Artificial Intelligence Act. Legal Issues in the Digital Age, 4(3), 97–116.

19. Melnikova, E. N. (2024). The problem of the “responsibility gap” for harm caused by artificial intelligence systems and applications. Russian Juridical Journal, 4. (In Russ.).

20. Novitsky, I. B. (1916). Goodwill principle in the draft law of obligations. Vestnik Grazhdanskogo Prava, 6–8. (In Russ.).

21. Orwat, C., Bareis, J., Folberth, A., Jahnel, J., & Wadephul, C. (2022). Normative Challenges of Risk Regulation of Artificial Intelligence and Automated Decision-Making. Karlsruhe. https://arxiv.org/pdf/2211.06203

22. Pearl, J. (2018). Theoretical Impediments to Machine Learning With Seven Sparks from the Causal Revolution. University of California*, Los Angeles.

23. Pearl, J., Glymour, M., & Jewell, N. (2019). Causal Inference in Statistics: A Primer A Primer Judea Pearl, Maria Glymour, and Nicholas Jewell, John Wiley & Sons, Ltd., Chichester, UK. Biometrics, 75(2), 708–709. https://doi.org/10.1111/biom.13079

24. Perrow, C. (1999). Normal Accidents: Living with High Risk Technologies. Princeton: Princeton University Press. https://archive.org/details/normalaccidentsl00perr

25. Pourkamali-Anaraki, F., Husseini, J. F., & Stapleton, S. E. (2024). Probabilistic Neural Networks (PNNs) for Modeling Aleatoric Uncertainty in Scientific Machine Learning. arXiv:2402.13945v1. https://arxiv.org/html/2402.13945v1

26. Rathkopf, C. (2026). From Hallucination to Reliability: Generative Modeling and the Structure of Scientific Inference. arXiv:2504.08526. https://arxiv.org/abs/2504.08526

27. Rudin, C. (2019). Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature Machine Intelligence, 1, 206–215. https://doi.org/10.1038/s42256-019-0048-x

28. Rudin, C., Chen, C., Chen, Z., Huang, H., Semenova, L., & Zhong, C. (2021). Interpretable Machine Learning: Fundamental Principles and 10 Grand Challenges. Duke University & University of Maine. https://arxiv.org/pdf/2103.11251

29. Russell, S. (2019). Human Compatible: Artificial Intelligence and the Problem of Control. New York: Penguin. https://books.google.at/books/about/Human_Compatible.html?id=M1eFDwAAQBAJ&redir_esc=y

30. Santoni de Sio, F., & Mecacci, G. (2021). Four Responsibility Gaps with Artificial Intelligence: Why they Matter and How to Address them. Philosophy & Technology, 34, 1057–1084. https://doi.org/10.1007/s13347-021-00450-x

31. Selivanova, E. S., & Sarkisyan, V. V. (2025). Legal models of liability for harm caused by AI systems: foundations, mechanisms, and formation criteria. Vestnik Yuridicheskogo Fakulteta Yuzhnogo Federalnogo Universiteta, 12(3), 68–77. (In Russ.).

32. Shirvindt, A. M. (2014). Good faith principle in the Russian Civil Code and comparative law. In Aequum ius. From friends and colleagues on the 50th anniversary of Prof. D. V. Dozhdev (p. 231). Moscow: Statut. (In Russ.).

33. Szegedy, C., Zaremba, W., Sutskever, I., Bruna, J., Erhan, D., Goodfellow, I., & Fergus, R. (2013). Intriguing properties of neural networks. https://arxiv.org/abs/1312.6199

34. Yeung, K. (2017). Algorithmic regulation: A critical interrogation. Regulation & Governance, 12(6). https://doi.org/10.1111/rego.12158

35. Zhou, L., Moreno-Casares, P. A., Martínez-Plumed, F., Burden, J., Burnell, R., Cheke, L., Ferri, C., Marcoci, A., Mehrbakhsh, B., Moros-Daval, Y., Ó hÉigeartaigh, S., Rutar, D., Schellaert, W., Voudouris, K., & Hernández-Orallo, J. (2023). Predictable Artificial Intelligence. arXiv:2310.06167. https://doi.org/10.48550/arXiv.2310.06167