Civil liability for harm caused by artificial intelligence systems has become an increasingly relevant issue as traditional liability models struggle to address autonomous, opaque, and unpredictable technologies. The challenge is that rules built around human conduct do not always fit systems whose harmful outputs may result from internal decision-making processes that are difficult to trace or explain.
Under Brazil’s fault-based liability regime, Article 186 of the Civil Code requires proof of intent or negligence. That standard can be difficult to apply when the harmful decision results from the internal logic of an AI system. Strict liability may offer an alternative, but it raises its own challenges, including how to identify the responsible party, whether the developer, manufacturer, or operator, and how to address defenses based on accident or unforeseeable events.
Against this background, proposals such as qualified strict liability, including a reversal of the burden of proof, and mandatory insurance for high-risk AI systems have gained traction.
Brazil still lacks specific legislation on the matter, leaving liability issues to judicial interpretation. Several bills are currently pending and seek to regulate the topic, particularly Bill (PL) 2338/2023.
PL 2338/2023 is Brazil’s main legislative proposal on the governance and regulation of artificial intelligence systems. It imposes strict liability on the supplier or operator of an AI system, requiring compensation for harm regardless of proof of fault. It also requires high-risk systems to maintain oversight mechanisms designed to prevent harm to fundamental rights.
Until specific legislation is enacted, one recommended approach is to apply the Brazilian Consumer Defense Code by analogy to AI systems offered as products or services. Under this framework, manufacturers, suppliers, and service providers may be held strictly liable for harm caused by defective products or services.
For other legal relationships, courts may turn to the risk-creation theory. Under this approach, anyone who develops or engages in an activity that creates risks for third parties may be held liable for resulting harm, even without proof of fault. Liability may also be imposed jointly and severally on the developer, data provider, and end user. As technology continues to evolve, the legal system must strike a workable balance between innovation and effective protection for victims.
Glossary:
Fault-based liability: A liability regime that requires proof of intent, negligence, recklessness, or other culpable conduct before compensation may be awarded.
Strict liability: A liability regime under which a party may be held responsible for harm regardless of proof of fault.
Article 186 of the Civil Code: A provision of the Brazilian Civil Code establishing liability for unlawful acts resulting from intentional or negligent conduct.
Reversal of the burden of proof: A procedural mechanism that shifts responsibility for proving certain facts from one party to another.
Qualified strict liability: A strict liability model that incorporates additional procedural safeguards or evidentiary mechanisms, such as a reversal of the burden of proof.
Bill (PL) 2338/2023: Brazil’s principal legislative proposal addressing the governance, regulation, and liability framework for artificial intelligence systems.
Presumption of causation: A legal mechanism that allows a causal link between conduct and harm to be presumed under certain circumstances unless evidence demonstrates otherwise.
Risk-creation theory: A legal doctrine under which a person or entity that creates risks for others through its activities may be held liable for resulting harm, even without proof of fault.
Brazilian Consumer Defense Code: Brazil’s primary consumer protection statute, governing consumer rights and supplier obligations.
