The question of liability for harm caused by artificial intelligence systems is increasingly prominent in legal and regulatory discussions across the United States, with implications extending to technology companies worldwide. As AI applications become more widespread, determining who should be held responsible when these systems malfunction or cause damage remains uncertain.
Historically, during the rise of the internet, U.S. lawmakers limited the liability of online platforms for user-generated content. In contrast, the current AI landscape has prompted debate over whether technology companies should bear more responsibility. For example, if a driver uses a self-driving feature and causes an accident, the driver’s responsibility is clear. However, the role of the AI system manufacturer is less defined—should the developer be liable even if the system was misused? Similarly, a Canadian court recently ruled that an airline must honor false policy information provided by its customer service chatbot, raising questions about corporate accountability for AI errors. U.S. courts are also considering cases alleging harm linked to AI chatbot interactions.
Legal experts remain divided on the appropriate approach. One key debate centers on whether to adopt “strict liability,” where manufacturers are held accountable for any harm caused by their products regardless of fault or intent. Proponents argue this would incentivize companies to prioritize safety. The European Union is progressing in this direction through revisions to its Product Liability Directive, which now explicitly includes software and AI systems. These changes make it easier for consumers to seek compensation by presuming defects if technical complexity obscures evidence or if companies fail to disclose necessary information. Liability could extend to failures such as neglecting cybersecurity updates.
Chinese companies that develop open-source AI models may find partial relief because the EU’s Directive exempts open-source developers from liability—except when their software is integrated into commercial products. This creates ambiguity, especially in cases where open-source AI is embedded in larger projects like smart city initiatives, potentially exposing developers to liability despite their indirect role.
Some experts warn that strict liability could hinder innovation. Yale law professor Kent Ramakrishnan argues that if companies are liable for all misuse of their AI models but capture only part of the societal benefits, they might withhold beneficial technologies altogether. Since harmful use cannot be fully prevented, imposing excessive liability risks discouraging development rather than promoting safety. He cites examples such as a novice programmer using AI to facilitate unauthorized computer access, or AI-enabled bulk emailing used to distribute malware, cases where holding the developer fully liable would be unreasonable.
Judicial responses to AI-related issues have so far tended to penalize users rather than developers. For instance, lawyers fined for submitting fabricated cases generated by chatbots have faced sanctions, but the AI companies that created the tools have not, partly because the technology was not marketed as legally authoritative.
As lawmakers and courts continue to examine these complex questions, a consensus on the appropriate legal framework for AI liability remains elusive. The evolving nature of AI technology and its varied applications present ongoing challenges to establishing clear rules for accountability.