TLDR:
AI liability addresses the legal allocation of responsibility when AI systems cause harm—to users, third parties, or property. The question is who bears legal responsibility along the value chain: the foundation model developer, the application developer, the deployer, the operator, or the user. Multiple regulatory frameworks are emerging to allocate this responsibility.
Liability Frameworks
Major frameworks include: traditional tort and product liability (negligence, strict liability for defective products), EU Product Liability Directive (revised 2024 to cover AI as a “product”), proposed EU AI Liability Directive (lowering evidentiary burden for AI-caused harm), professional negligence (when professionals use AI in their work—lawyers, doctors, financial advisors), employment law obligations for AI-mediated decisions, contract-based liability (allocation between AI vendor and customer), and increasingly insurance products covering AI-specific risks.
Allocation Across the Value Chain
The complexity of modern AI value chains makes liability allocation difficult. A typical deployment involves: a foundation model provider (Anthropic, OpenAI), a fine-tuning or RAG layer (the application developer), a deployer integrating into a workflow, and end users. The EU AI Act and AI Liability Directive attempt to allocate responsibility primarily to the provider of high-risk AI systems and the deployer, with specific obligations for general-purpose AI. Contract negotiations between these parties increasingly focus on warranty disclaimers, indemnification, and liability caps.
Practical Risk Management
Companies should implement: clear contractual allocation of AI risk between vendors and customers, professional liability and product liability insurance covering AI-specific scenarios, documentation of AI decision-making processes for high-stakes use cases, human oversight requirements for legally consequential decisions, and incident response plans for AI failures. Litigation in AI areas is increasing—particularly around discrimination (employment AI), hallucinations (legal/medical professional liability), and copyright (training data). Founders deploying AI in regulated industries should engage specialized counsel early.