It is important, first and foremost, to recognize that the classical models of civil liability, as set forth in the Portuguese Civil Code, are based on an anthropocentric framework: they presuppose the existence of human conduct, attributable to an agent, who acts with intent or negligence. AI disrupts this paradigm, insofar as it enables automated decision-making based on machine learning processes, whose internal logic is not always comprehensible—the so-called “black box” problem.
In this context, a key initial question concerns the attribution of liability. Who is liable for damages caused by AI systems? The programmer? The manufacturer? The professional user? Or, in certain circumstances, the beneficiary of the automated decision themselves? The answer is not straightforward and tends to depend on the specific value chain of the AI system and the degree of control exercised by each party involved.
In the context of subjective civil liability, proving fault becomes particularly complex. Indeed, decisions must be capable of being adequately explained and understood in light of the specific case, rather than based on the algorithm’s predictions (black box), which may raise issues related to damages resulting from AI-based decisions, such as the burden of proof for fault and the causal link. This evidentiary difficulty can, in practice, weaken the injured party’s position.
For this reason, the use of strict or presumptive liability mechanisms has been gaining prominence. At the European level, the most recent legislative initiatives aim to facilitate the burden of proof for the injured party and to potentially establish presumptions of causation in certain circumstances (for example, when there is a breach of transparency or security obligations), and it is expected that such guidelines will also be incorporated into the Portuguese legal system.
Another critical issue concerns the causal link. In AI systems, harm can result from a multitude of factors: biased training data, programming errors, decisions made by the system itself, or even interactions with other systems. Identifying the relevant causal factor can prove extremely difficult, especially when AI operates in a dynamic and adaptive manner. This reality challenges the traditional criteria of adequate causation theory and may justify more flexible solutions or even the reversal of the burden of proof in certain contexts.
In the insurance sector, these challenges are particularly acute. On the one hand, AI is used for underwriting risks, claims management, and fraud prevention; on the other hand, it is also a source of new risks that may give rise to civil liability. For example, an AI system that wrongfully denies a claim based on discriminatory or erroneous criteria may result in the insurer’s liability, both contractually and non-contractually.
In addition, there is the question of whether existing insurance products are adequate to cover risks associated with AI. Professional liability or product liability insurance may not be sufficient to cover all of these risks, especially when intangible damages are involved, such as algorithmic discrimination or violations of fundamental rights. It is therefore foreseeable that specific insurance products for AI risks will be developed, and that existing clauses will be adapted.
At the same time, AI regulation emphasizes the importance of explainability and human oversight (“human in the loop”). These requirements, in addition to aiming to protect users’ rights, also have direct implications for civil liability: the existence of human control mechanisms can facilitate the attribution of liability and the demonstration of fault, while their absence may reinforce the need for strict liability regimes.
In conclusion, artificial intelligence poses structural challenges to the traditional civil liability regime, requiring an adaptation of its underlying principles and instruments. The trend points toward greater objectification of liability, the facilitation of proof by the injured party, and the development of specific solutions at the regulatory and insurance levels. For legal practitioners and the insurance sector, keeping pace with these developments will be crucial, not only for risk management but also for building an environment of trust in the use of AI.
Belzuz Abogados, S.L.P. – Portugal Branch has a multidisciplinary team with extensive experience advising domestic and foreign insurance companies and insurance brokers, and can provide significant assistance in monitoring and implementing new legislative developments, particularly those related to AI.