Should artificial intelligence have legal personality?

Artificial intelligence has developed extraordinarily fast in recent years. Industry leaders and several researchers predict that artificial general intelligence (AGI), able to learn and adapt at a human level, may arrive within a few years. Even if such forecasts are optimistic, autonomous systems already make decisions that humans may not fully foresee or control. When such a system causes harm, who is liable? And is a legal system that recognises only natural and legal persons ready for this? This post is based on a 2026 article in Juridica by Olavi-Jüri Luik, Kadi Saluste and Mats Volberg.¹

The problem: a liability gap

The basic principle of civil liability is that the person who caused the consequence, or who assumed the risk of it, is liable. That model assumes AI is a tool that always acts in someone's interest and under someone's control.

The article distinguishes two properties of AI. An autonomous system can set its own goals. An independent system can choose its own means to reach a given goal. Both can produce situations in which no human caused the harm or knowingly assumed its risk.

Two thought experiments illustrate the problem:

1. A virtual assistant built to purchase goods for a company starts, on its own initiative, to conclude consumer credit agreements or uses impermissible tactics in negotiations.
2. A park cleaning robot decides that to prevent litter it must stop birds and people from entering the park, and acts accordingly.

In such situations it is hard to argue that the manufacturer or user caused the consequence or assumed its risk where the machine itself departed from its original state.

Current law and the EU position

Estonian law recognises two kinds of person: natural persons (General Part of the Civil Code Act (tsiviilseadustiku üldosa seadus, TsÜS) § 7 et seq.) and legal persons (TsÜS § 24 et seq.).² Estonian legal literature has mostly held that giving AI personhood is neither necessary nor sensible: AI is a thing, and its manufacturer or keeper is liable.

The EU takes the same approach. The AI Act (Regulation (EU) 2024/1689) and the new Product Liability Directive (Directive (EU) 2024/2853) treat AI as a product for which people and organisations are responsible.³ In its 2020 resolution on civil liability for AI, the European Parliament considered that giving AI systems legal personality was not necessary. The European Commission withdrew its proposed AI Liability Directive in 2025.

Legal personality is not fixed

Historically, being a person in law has not always coincided with being human. In Roman law slaves were not legal subjects, and full legal capacity belonged only to Roman citizens. The legal person is likewise a creation of law, made for economic reasons.

In recent decades legal personality has also been given to natural features. In 2017 New Zealand recognised the Whanganui River by statute as a legal person represented by statutory guardians. In the same year the High Court of Uttarakhand in India declared the Ganges and Yamuna rivers legal persons (20 March 2017), but on 7 July 2017 the Supreme Court of India stayed that ruling. These examples show that legal personality is a policy choice, not a law of nature.

The authors' position: for an electronic person

The authors support legal personification of autonomous and independent AI, for these reasons:

• A clear liability framework. Personification allows rights and duties to be attributed to AI and claims to be directed at it, filling the gap that arises where traditional models fail.
• Legal certainty for victims and developers. A predictable framework supports both injured parties and innovation.
• Regulating forced shutdown. Personification makes it possible to think through and regulate how a dangerous AI's activity is ended, which is considered important for mitigating existential risk.
• International coordination. AI is not one country's internal problem. Personification requires a common international approach, for example model principles developed at UN level.

At the same time, the authors firmly oppose giving AI human or civil rights. Software can copy itself practically without limit, so giving it, for example, the right to vote would make human self-determination meaningless. Legal personality would thus mean limited and specific legal capacity, mainly to hold property rights and obligations and bear liability, not moral status.

Counter-arguments

The article also weighs counter-arguments that deserve serious attention:

• Diffusion of responsibility. If AI itself is liable, manufacturers and users may hide behind it. The electronic person's assets may be negligible, leaving the victim uncompensated.
• Fault and punishment. AI lacks self-awareness, so criminal liability makes little sense because punishment cannot serve its purpose. In civil law the problem is milder, since risk allocation mechanisms are available (strict liability for sources of greater danger, attribution through legal persons).
• Technical definition. It is hard to define legally which system is autonomous enough to justify personification.

What this means for insurance

From an insurance law perspective, the electronic person is a logical extension of the legal person liability model. If AI bears liability, it must have assets or insurance from which to pay. Many electronic person models therefore centre on compulsory liability insurance or a compensation fund, similar to motor insurance, so that victims are protected even if the electronic person's own assets are small.

For insurers this would create a new insured subject and a new risk class for which no statistics yet exist. In the near term it is more realistic that AI risks remain covered through existing products: product liability, professional indemnity, cyber and D&O insurance. Insurers already ask about AI use in underwriting, and some policies contain AI-related exclusions or limitations, so policyholders should read wordings carefully (see AI liability and autonomous vehicles).

Three time horizons

AI can be discussed on three time horizons: the actual position today, the likely near future and a more speculative distant future. The article focuses on the middle level, systems that are autonomous, independent or both.

Today most AI systems are tools: they perform tasks set by humans and their conduct is attributable to the user or manufacturer. Existing law suffices at this stage: contract law, product liability and strict liability for sources of greater danger.

In the near future, systems may spread that act independently for longer periods and interact with other systems without human intervention: AI agents that conclude contracts, manage assets and deal with customers. This is where existing law gives uncertain answers. Is an agent's declaration of intent attributable to the user if the user did not foresee it? Is it force majeure if the system behaved unforeseeably?

Distant future scenarios in which AI surpasses humans in all fields are outside the article's scope. The authors nevertheless stress that anticipatory legislation is needed, because reacting to consequences may come too late.

Systems that act directly in the physical world (self-driving cars, robots) also differ in risk profile from purely digital systems (virtual assistants). The former can cause personal injury, the latter mainly financial loss, but on a large scale and quickly. A liability model must reflect that difference.

Conclusion

Whether AI should have legal personality is no longer science fiction. Current Estonian and EU law treats AI as a product, but the development of autonomous and independent systems may create liability gaps that traditional models do not cover. The authors consider anticipatory legislation justified and support an electronic person with limited legal capacity, not the extension of human rights to AI. Any such model depends on insurance or a fund that guarantees compensation for victims.

Based on: O.-J. Luik, K. Saluste, M. Volberg, "Kas tehisarule tuleks anda õigussubjektsus?", Juridica 2026/4, pp. 331–341. Rewritten for the web.

Related pages

• AI liability and autonomous vehicles: will insurance replace tort law?
• Product liability in Estonia
• Professional indemnity in Estonia
• Publications

References