Rethinking Liability, Accountability and Human Oversight

1. Introduction

Artificial intelligence (AI) is increasingly influencing decisions and activities that can directly affect individuals, businesses and society. Although AI systems offer significant benefits, their use may also cause tangible harm, including financial loss, discrimination, privacy violations and unsafe outcomes. These risks raise a fundamental legal question: who should bear responsibility when an AI system causes damage? Traditional liability principles generally rely on identifying a responsible person, establishing a relevant duty or fault, and demonstrating a causal connection between conduct and harm. However, AI systems may involve several actors, including developers, providers and deployers, whose respective contributions are difficult to distinguish. International standards therefore emphasise accountability, transparency and meaningful human oversight.[1]

This article examines whether existing liability principles can adequately address AI-related harm and argues that responsibility should remain attributable to the human or legal persons involved, according to their roles, control and failures to meet applicable duties.[2]

2. The Legal Challenge of AI-Generated Harm

The central difficulty in establishing liability for AI-related harm is the complexity of determining how damage occurred and which actor should legally answer for it. Unlike conventional products, certain AI systems can operate with varying degrees of autonomy, adapt through data and produce outcomes that may not have been specifically anticipated by their developers or users. This can make it difficult for an injured party to identify the precise cause of harm or prove negligence.[3]

Moreover, responsibility may be distributed across several participants in the AI lifecycle. A developer may design the system, a provider may place it on the market, and a deployer may use it in a particular operational environment. Each may influence the outcome differently. Therefore, assigning liability solely to the user or developer may overlook other relevant contributions.[4]

Another challenge concerns evidence and transparency. Where an AI system operates through complex models or limited explainability, affected individuals may struggle to understand the decision that harmed them or obtain the information needed to establish causation.[5] Consequently, the legal challenge is not simply identifying who caused the harm, but ensuring that existing liability rules provide an effective route to accountability and compensation.

3. Identifying Liability Across the AI Lifecycle

Liability for AI-related harm should be assessed by examining the responsibilities of each actor throughout the system’s lifecycle, rather than automatically attributing the damage to one participant. Developers may bear responsibility where defective design, inadequate testing or failure to address foreseeable risks contributes to harm. Their obligations should include taking reasonable precautions, documenting system limitations and addressing identified vulnerabilities before deployment.[6]

Providers may also be responsible where they fail to supply appropriate instructions, disclose relevant risks or maintain required safeguards. Under the European Union Artificial Intelligence Act, providers of high-risk AI systems are subject to specific obligations concerning risk management, technical documentation, transparency and human oversight. These requirements demonstrate how regulatory duties can help establish standards against which conduct may be assessed.[7]

Deployers and professional users must also be considered. They may incur responsibility when they use systems outside their intended purpose, ignore warnings, provide unsuitable input data or fail to exercise appropriate human supervision. The applicable legal test should therefore consider each actor’s actual role, degree of control, knowledge of foreseeable risks and compliance with relevant duties.[8]

4. Human Oversight, Transparency and Accountability

Human oversight is essential to ensuring that AI systems remain subject to meaningful human control, particularly when their outputs may affect fundamental rights, safety or financial interests. Effective supervision requires more than the formal presence of a human operator; individuals responsible for oversight must understand the system’s capabilities and limitations, interpret its outputs and possess the authority to reject, override or interrupt decisions when necessary.[9]

Transparency is equally important because accountability becomes difficult when affected individuals cannot understand how an AI system reached a particular outcome. Providers should therefore communicate relevant system limitations, foreseeable risks and operational requirements to deployers. Appropriate documentation and record-keeping can also help investigators and courts reconstruct the circumstances surrounding an incident and assess whether the relevant actors fulfilled their legal duties.[10]

Accountability must also include accessible mechanisms for complaints, investigation and redress. A person harmed by an AI-assisted decision should not be left without an effective remedy merely because the system’s internal operation is technically complex. Human oversight, transparency and traceability should therefore operate together: they help prevent harm, identify failures and provide evidence when legal responsibility is disputed.[11]

5. The Egyptian and International Legal Framework

In Egypt, Article 163 of the Civil Code provides a general basis for compensation where wrongful conduct causes damage. It may apply to certain AI-related harms if the claimant establishes fault, damage and causation. However, identifying responsibility may be difficult when several actors contribute to the harmful outcome.[12]

Egypt’s Personal Data Protection Law No. 151 of 2020 is also relevant where AI-related harm involves the unlawful processing or misuse of personal data. The law regulates electronically processed personal data and establishes obligations for relevant data controllers and processors. Nevertheless, it should not be interpreted as a comprehensive framework governing every form of AI-related harm, since its scope is principally data protection.[13]

At the policy level, Egypt has adopted its National AI Strategy for 2025–2030 and published a national AI governance framework in 2026. These initiatives promote responsible, human-centred AI development, but policy and guidance instruments should be distinguished from binding legislation.[14]

Internationally, the OECD AI Principles and UNESCO Recommendation on the Ethics of AI provide important guidance on accountability, transparency and human oversight. Although they are not equivalent to binding liability statutes, they offer useful standards for developing responsible national approaches.[15]

6. Conclusion

AI-related harm requires a liability approach that reflects the responsibilities of all actors involved in the AI lifecycle. National legal frameworks should clarify the duties of developers, providers and deployers, strengthen transparency and documentation requirements, and facilitate access to evidence for affected individuals. Human oversight must remain meaningful, particularly where AI systems influence decisions affecting rights and safety. Ultimately, artificial intelligence should not become a legal shield behind which responsible parties can escape accountability. The law must ensure that technological progress develops alongside responsibility, fairness and effective protection for those harmed.