Liability for harm caused by AI in healthcare: an overview of the core legal concepts
Open accessThe DOI matches its Crossref record; 33 of the 38 references it cites were found in a registry or an index; 27 of the 80 sources its claims cite were checked for support.
Abstract
The integration of artificial intelligence (AI) into healthcare in Africa presents transformative opportunities but also raises profound legal challenges, especially concerning liability. As AI becomes more autonomous, determining who or what is responsible when things go wrong becomes ambiguous. This article aims to review the legal concepts relevant to the issue of liability for harm caused by AI in healthcare. While some suggest attributing legal personhood to AI as a potential solution, the feasibility of this remains controversial. The principal–agent relationship, where the physician is held responsible for AI decisions, risks reducing the adoption of AI tools due to potential liabilities. Similarly, using product law to establish liability is problematic because of the dynamic learning nature of AI, which deviates from static products. This fluidity complicates traditional definitions of product defects and, by extension, where responsibility lies. Exploring alternatives, risk-based determinations of liability, which focus on potential hazards rather than on specific fault assignments, emerges as a potential pathway. However, these, too, present challenges in assigning accountability. Strict liability has been proposed as another avenue. It can simplify the compensation process for victims by focusing on the harm rather than on the fault. Yet, concerns arise over the economic impact on stakeholders, the potential for unjust reputational damage, and the feasibility of a global application. Instead of approaches based on liability, reconciliation holds much promise to facilitate regulatory sandboxes. In conclusion, while the integration of AI systems into healthcare holds vast potential, it necessitates a re-evaluation of our legal frameworks. The central challenge is how to adapt traditional concepts of liability to the novel and unpredictable nature of AI—or to move away from liability towards reconciliation. Future discussions and research must navigate these complex waters and seek solutions that ensure both progress and protection.
Overview
CoChat Summary
TL;DR
This article reviews core legal concepts relevant to liability for harm caused by AI in healthcare, with a focus on Africa. It argues that traditional liability approaches face serious challenges with AI, and that reconciliation through a sui generis dispute-resolution institution may better support a regulatory sandbox.
Key findings
- AI personhood is controversial and currently unlikely for existing systems; it would not solve compensation because AI systems cannot own assets, and insurance alternatives raise administrative-cost and risk-assessment concerns.
- Extending principal–agent or “human-in-the-loop” rules would make physicians responsible for AI-assisted decisions, which may disincentivise AI adoption and leaves unclear how AI fits into the standard of care.
- Product liability is complicated by AI’s dynamic learning: Townsend et al. (2023) found that eleven out of twelve African countries surveyed provide strict liability for harmful or defective goods, but proving an AI “defect,” causation, and developer responsibility is difficult.
- Fault-based liability struggles with causation and fault, while strict liability eases compensation but raises economic, reputational, and cross-jurisdictional concerns; reconciliation is proposed as a promising alternative but lacks comparable scholarship.
Methods
- Design: narrative legal overview/review of legal concepts relevant to liability for AI harm in healthcare, aimed at setting the stage for legal development and policy initiatives in Africa.
- Data/sample: no empirical sample; the article draws on legal literature, policy documents, and examples of AI use in African healthcare, including South Africa, Ghana, Uganda, and Nigeria.
- Analysis: conceptual and doctrinal discussion of approaches to attributing liability: AI personhood, principal–agent relationship, product liability, fault-based remedies, strict liability, and reconciliation.
Limitations
- The paper states that reconciliation in the context of AI-driven harm in healthcare lacks the same level of scholarship as liability-based approaches and deserves more academic attention.
- The paper states that there is currently no case law to guide fault-based liability principles, particularly where AI has an unknown flaw that was not reasonably foreseeable.
- Evident from the text: as an overview, the article does not empirically test the proposed approaches or provide primary data, relying instead on cited literature and illustrative examples.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
25 Valid · 3 Not valid · 50 Not checked · 55 Own results and interpretations
At least one cited source doesn’t support this claim or points to a different paper, and none confirmed it — see the verdicts below.
In the paper
“Current norms suggest that claims for damages will favour standard care pathways, even where AI systems recommend non-standard treatments (Tobia et al., 2021).”
§ Liability for harm caused by AI in healthcare: an overview of the core legal concepts > 5 Approaches to attributing liability > 5.1 Principal–agent relationship
What the sources say
- Disputedcites Tobia et al., 2021 sourceExperimental — accuracy not yet measured
The Results (page unavailable) report higher reasonableness ratings when physicians accepted a nonstandard AI recommendation than when they rejected it to provide standard care. The Discussion (page unavailable) likewise says rejecting nonstandard AI advice does not necessarily make physicians safer from liability. The study assessed lay liability judgments, not damages awards. (LLM, conf=0.88)
Checked by a reasoning model — may be wrong.
hedgedstates a possible 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.
- Disputed
Every source the paper cites for this claim was checked and supports it.
In the paper
“An AI system fulfilling roles normally delegated to humans does not mean that personhood necessarily follows (Thaldar and Naidoo, 2021).”
§ Liability for harm caused by AI in healthcare: an overview of the core legal concepts > 4 Challenging our understanding of liability: AI and personhood
What the sources say
- Supportedcites Thaldar and Naidoo, 2021 source
In “Does this mean that AI has personhood?”, the paper says that allowing AI inventorship does not necessarily imply personhood. It explains that exhibiting inventiveness, a characteristic associated with humans, does not make AI legally a person. (LLM, conf=0.93)
plain- 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.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“Townsend et al. (2023) found that eleven out of twelve African countries surveyed provide for strict liability of harmful or defective goods in their consumer protection laws.”
§ Liability for harm caused by AI in healthcare: an overview of the core legal concepts > 5 Approaches to attributing liability > 5.2 Product liability
What the sources say
- Supportedcites Townsend et al. (2023) source
The Analysis section states that eleven of the twelve countries studied provide for strict product liability for harmful or defective goods under their consumer protection regimes. It identifies Cameroon as the exception, where liability is negligence-based. (LLM, conf=0.95)
plain- 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.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“In Europe, the “developmental risk defence” allows a producer to avoid liability on the basis that scientific knowledge at the time of production was unable to detect the existence of a defect in the product (Holm et al., 2021).”
§ Liability for harm caused by AI in healthcare: an overview of the core legal concepts > 5 Approaches to attributing liability > 5.2 Product liability
What the sources say
- Partly supportedcites Holm et al., 2021 sourceExperimental — accuracy not yet measured
In “Possible Solutions within a Fault-Based System” (page not provided), the paper quotes a UK Consumer Protection Act development defence based on whether scientific and technical knowledge could have enabled a producer to discover a defect while the product was under its control. That supports the substance of the defence, but the paper identifies a UK provision rather than establishing that it applies across Europe; it also does not specify the time of production as the relevant point. (LLM, conf=0.84)
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.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“No-fault systems also separate the compensation and liability claims (Holm et al., 2021).”
§ Liability for harm caused by AI in healthcare: an overview of the core legal concepts > 5 Approaches to attributing liability > 5.4 Strict liability
What the sources say
- Supportedcites Holm et al., 2021 source
In “Fault and No-Fault Compensation for Medical Error,” the paper says compensation does not require litigation or showing that anyone was at fault. “No-Fault System?” likewise says a patient can be compensated without locating responsibility or negligence. This supports the paraphrase that compensation is separated from liability. (LLM, conf=0.91)
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
35 more claims · 6 with their cited sources checked
Create a free account to see all 133 claims and every source verdict.
- Probably not a substantial claim. 79
- It may state two separate claims as one. 38
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 cited source says what the claim attributes to it.
In the citing paper
“Healthcare technology applications recognize intrinsic tensions between technological advancement and legal accountability when AI systems serve as intermediaries of conclusions and outcomes (Bottomley & Thaldar, 2023).”
From Causation in AI Tort Litigation: Legal Dilemmas of Algorithmic Black Boxes and Burden of Proof Allocation (2025) · cites this paper as Bottomley & Thaldar, 2023
Citations
Cites
Floridi, L., Cowls, J., Beltrametti, M., Chatila, R., Chazerand, P., Dignum, V., Luetge, C., Madelin, R., Pagallo, U., Rossi, F., et al. (2018) · Minds and Machines · doi:10.1007/s11023-018-9482-5 · Cited by 4,774 · cited 1×
“This trust will be important for the future of AI as a lack of trust could permanently harm the reputation of AI in healthcare, or lead to additional costs through inef fi cient regulation or repeated amendment (Floridi et al., 2018).”
BackgroundAli, S., Abuhmed, T., El-Sappagh, S., Muhammad, K., Alonso-Moral, J. M., Confalonieri, R., Guidotti, R., Del Ser, J., Díaz-Rodríguez, N., & Herrera, F. (2023) · Information Fusion · doi:10.1016/j.inffus.2023.101805 · Cited by 1,762
Ethical and legal challenges of artificial intelligence-driven healthcare
Gerke, S., Minssen, T., & Cohen, G. (2020) · Artificial Intelligence in Healthcare · doi:10.1016/b978-0-12-818438-7.00012-5 · Cited by 1,170
The ethics of artificial intelligence
Bostrom, N., & Yudkowsky, E. (2014) · The Cambridge Handbook of Artificial Intelligence · doi:10.1017/cbo9781139046855.020 · Cited by 918
Potential liability for physicians using artificial intelligence
Price, W. N., Gerke, S., & Cohen, I. G. (2019) · JAMA · doi:10.1001/jama.2019.15064 · Cited by 579
Global evolution of research in artificial intelligence in health and medicine: a bibliometric study
Tran, B. X., Vu, G. T., Ha, G. H., Vuong, Q.-H., Ho, M.-T., Vuong, T.-T., La, V.-P., Ho, M.-T., Nghiem, K.-C. P., Nguyen, H. L. T., et al. (2019) · Journal of Clinical Medicine · doi:10.3390/jcm8030360 · Cited by 472 · cited 1×
“…in healthcare systems (Joshi,and Morley, 2019), Africa has had relatively little meaningful interaction with AI in healthcare both academically (Tran et al., 2019) and clinically (Owoyemi et al., 2020) and, currently, African countries are at a nascent stage in their AI regulatory policies…”
BackgroundRobotics and the Lessons of Cyberlaw
Calo, R. (2014) · doi:10.2139/ssrn.2402972 · Cited by 269 · cited 2×
“The nature of emerging technologies is that we need time to understand them and develop policies and regulations which will encourage equitable use (Calo, 2015).”
BackgroundArtificial Intelligence for healthcare in africa
Owoyemi, A., Owoyemi, J., Osiyemi, A., & Boyd, A. (2020) · Frontiers in Digital Health · doi:10.3389/fdgth.2020.00006 · Cited by 199 · cited 4×
“In Nigeria, Ubenwa is using AI to improve the diagnosis of birth asphyxia in low-resource settings (Owoyemi et al., 2020).”
BackgroundMethodInfluentialSocial and juristic challenges of artificial intelligence
Perc, M., Ozer, M., & Hojnik, J. (2019) · Palgrave Communications · doi:10.1057/s41599-019-0278-x · Cited by 185 · cited 2×
“Of course, as Perc et al. (2019) consider, another approach may be to leave such decisions for the AI system to decide itself, or simply leave it to chance.”
BackgroundArtificial intelligence : a European perspective
European Commission. Joint Research Centre (2018) · EUR (Luxembourg) · doi:10.2760/936974 · Cited by 111
Artificial Intelligence and Legal Liability
Kingston, J. K. C. (2016) · Research and Development in Intelligent Systems XXXIII · doi:10.1007/978-3-319-47175-4_20 · Cited by 103 · cited 1×
“Furthermore, similar to criticism of AI personhood, critics of HITL argue that there is a dif fi culty in determining the correct standard against which to compare the conduct of the AI system (Kingston, 2016).”
BackgroundWhen does physician use of AI increase liability?
Tobia, K., Nielsen, A., & Stremitzer, A. (2021) · Journal of Nuclear Medicine · doi:10.2967/jnumed.120.256032 · Cited by 103 · cited 1×
“Current norms suggest that claims for damages will favour standard care pathways, even where AI systems recommend non-standard treatments (Tobia et al., 2021).”
BackgroundDignum, V. (2017) · arXiv · doi:10.48550/arxiv.1706.02513 · Cited by 103
Mapping the regulatory landscape of AI in healthcare in Africa
Townsend, B. A., Sihlahla, I., Naidoo, M., Naidoo, S., Donnelly, D.-L., & Thaldar, D. W. (2023) · Frontiers in Pharmacology · doi:10.3389/fphar.2023.1214422 · Cited by 76
Self-citationArtificial intelligence in clinical decision-making: Rethinking liability
Smith, H., & Fotheringham, K. (2020) · Medical Law International · doi:10.1177/0968533220945766 · Cited by 73
Is South Africa on the verge of a medical malpractice litigation storm
(2011) · UpSpace Institutional Repository (University of Pretoria) · doi:10.7196/sajbl.115 · Cited by 47 · cited 1×
“Furthermore, these systems can increase the reach of initiatives, while supplementing an already overburdened sector (Pepper and Slabbert, 2011).”
BackgroundArtificial intelligence in healthcare: Proposals for policy development in South Africa
Naidoo, S., Bottomley, D., Naidoo, M., Donnelly, D., & Thaldar, D. W. (2022) · South African Journal of Bioethics and Law · doi:10.7196/sajbl.2022.v15i1.797 · Cited by 40
Self-citationTowards shaping the future of responsible AI in africa
Eke, D. O., Chintu, S. S., & Wakunuma, K. (2023) · Social and Cultural Studies of Robots and AI · doi:10.1007/978-3-031-08215-3_8 · Cited by 33
Holm, S., Stanton, C., & Bartlett, B. (2021) · Health Care Analysis · doi:10.1007/s10728-021-00430-4 · Cited by 32 · cited 6×
“No-fault systems also separate the compensation and liability claims (Holm et al., 2021).”
BackgroundInfluentialPunishing artificial intelligence: legal fiction or science fiction
Abbott, R. B., & Sarch, A. F. (2019) · SSRN Electronic Journal · doi:10.2139/ssrn.3327485 · Cited by 31
BERTOLINI, A., & EPISCOPO, F. (2021) · European Journal of Risk Regulation · doi:10.1017/err.2021.30 · Cited by 31 · cited 2×
“Furthermore, there is a lack of guidance on the value of AI insurance policies as there are no standards against which to assess risk or begin a cost analysis (Bertolini and Episcopo, 2021).”
BackgroundLegal Personhood of Artificial Intelligence
Luzan, T. (2018) · Työväentutkimus Vuosikirja · Cited by 27
Whose Robot Is It Anyway?: Liability for Artificial-Intelligence-Based RobotsPreprint
Rachum-Twaig, O. (2019) · SSRN Electronic Journal · Cited by 26
Bashayreh, M., Sibai, F. N., & Tabbara, A. (2021) · Information & Communications Technology Law · doi:10.1080/13600834.2020.1856025 · Cited by 26 · cited 9×
“An extension of this idea is a risk-sharing approach (Bashayreh et al., 2021).”
BackgroundInfluentialIn support of “no-fault” civil liability rules for artificial intelligence
Marchisio, E. (2021) · SN Social Sciences · doi:10.1007/s43545-020-00043-z · Cited by 25 · cited 5×
“This has led some scholars to suggest that a mixture of fault and no-fault rules could provide equitable AI regulation (Marchisio, 2021).”
BackgroundInfluentialOn legal personhood of artificial intelligence
Miščević, N., & Savčić, S. (2024) · Zbornik radova Pravnog fakulteta Novi Sad · doi:10.5937/zrpfns58-50186 · Cited by 23
EU liability rules for the age of artificial intelligence
Buiten, M., de Streel, A., & Peitz, M. (2021) · SSRN Electronic Journal · doi:10.2139/ssrn.3817520 · Cited by 20 · cited 3×
“Fault-based liability is an important means of deterrence (Buiten et al., 2021).”
BackgroundInfluential(2017) · SSRN Electronic Journal · doi:10.38046/apjhle.2018.11.2.004 · Cited by 17
Sihlahla, I., Donnelly, D., Townsend, B., & Thaldar, D. (2025) · Developing World Bioethics · doi:10.1111/dewb.12436 · Cited by 16
Self-citationAI inventorship: the right decision?
Thaldar, D., & Naidoo, M. (2021) · South African Journal of Science · doi:10.17159/sajs.2021/12509 · Cited by 9
Self-citationDonnelly, D.-L. (2022) · Potchefstroom Electronic Law Journal · doi:10.17159/1727-3781/2022/v25i0a11118 · Cited by 8 · cited 3×
“Furthermore, establishing vicarious liability would be complicated as, currently, there is no means of determining whether the AI system “ acted negligently ” or what degree of control a medical practitioner should exert over an AI system (Donnelly, 2022).”
Background(2020) · Tydskrif vir die suid-afrikaanse reg · Cited by 2
(2021) · eYLS (Yale Law School)
Research works printed but not linked: 11
- 2018, Access partnership
- 2019, Liability for Artificial Intelligence and other emerging digital technologies
- 2019, Ethics guidelines for trustworthy AI
- Joshi, 2019, Artificial Intelligence: how to get it right. Putting policy into practice for safe data-driven innovation in health and care
- Jumper, 2020, Computational predictions of protein structures associated with COVID-19, DeepMind
- Lövtrup, 2020, In brief: artificial Intelligence in healthcare, Swed. Counc. Med. Ethics, 2
- 2021, Ltd and another v wentzel (1272/2019) [2021] ZASCA 40 (13 april 2021)
- Mukheibir, 2010, The law of delict in South Africa
- Sallstrom, 2019, Artificial intelligence in Africa’s healthcare: ethical considerations, Observer Res. Found. Issue Brief, 312, 12
- Singh, 2020, AI and data in South Africa’s health sector, Policy Action Netw., 6
- 2021, Ethics and governance of artificial intelligence for health: WHO guidance
Cited by
Parvin, N., Joo, S., Jung, J. H., & Mandal, T. K. (2025) · Nanomaterials · doi:10.3390/nano15120895 · Cited by 151
“This ambiguity can hinder efforts to address errors and learn from them [170].”
Artificial intelligence in nanotechnology for treatment of diseases
Heydari, S., Masoumi, N., Esmaeeli, E., Ayyoubzadeh, S. M., Ghorbani‐Bidkorbeh, F., & Ahmadi, M. (2024) · Journal of drug targeting · doi:10.1080/1061186x.2024.2393417 · Cited by 106
“Additionally, defining responsibility for errors caused by Ai systems is complex and requires clear liability protocols [110].”
Medical Hallucinations in Foundation Models and Their Impact on Healthcare
Kim, Y., Jeong, H., Chen, S., Li, S. S., Park, C., Lu, M., Alhamoud, K., Mun, J., Grau, C., Jung, M., et al. (2025) · arXiv · doi:10.48550/arxiv.2503.05777 · Cited by 33
Responsible artificial intelligence in Africa: towards policy learning
Plantinga, P., Shilongo, K., Mudongo, O., Umubyeyi, A., Gastrow, M., & Razzano, G. (2024) · Data & Policy · doi:10.1017/dap.2024.60 · Cited by 67
“…health sector was not included in our analysis, but there is a growing body of Africa-centric research around AI governance that is likely to provide useful insights around specific issues, such as accountability and liability, which could be relevant to other sectors (Bottomley and Thaldar, 2023).”
Sablone, S., Bellino, M., Cardinale, A. N., Esposito, M., Sessa, F., & Salerno, M. (2024) · Frontiers in Medicine · doi:10.3389/fmed.2024.1343456 · Cited by 42
Metadata
- Authors
- Bottomley, D., & Thaldar, D.
- Affiliations
- University of KwaZulu-Natal (ZA)
- Corresponding author
- Thaldar, Donrich
- Venue
- Frontiers in Pharmacology
- Publisher
- Frontiers Media SA
- Published
- 2023-12-14
- Volume
- 14
- Pages
- 1297353-1297353
- ISSN
- 1663-9812 (ISSN-L)
- PMC
- PMC10755877
- OpenAlex
- W4389732527
- Cited by
- 80
- References
- 38
- Citations by year
- 2026: 40 · 2025: 28 · 2024: 12
- FWCI
- 9.89
- Type
- journal article
- Language
- en
- Update notices
- corrigendum (2024-03-14)
- Open access
- gold
- doi.org
- Licence
- Locations
- Frontiers in Pharmacology (publishedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Funders
- National Institutes of Health
- National Institute of Mental Health (U01MH127690)
- Topics
- Social Sciences › Law › Digitalization, Law, and Regulation
- Social Sciences › Law › Medical Malpractice and Liability Issues
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Keywords
- artificial intelligence in healthcare, legal personhood of AI, principal-agent relationship, product liability, defective products, risk-based liability, strict liability, regulatory sandboxes, reconciliation, accountability, Africa
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 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
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
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.
