AI liability in 2026 spans contract, tort, product liability, and professional negligence. The revised EU Product Liability Directive (Directive (EU) 2024/2853) now explicitly covers software and AI, while the AI Liability Directive proposal was withdrawn from the Commission's 2025 work programme — leaving national tort law to fill the gap.
AI liability concerns who pays when AI causes harm. Harm types include bodily injury, property damage, pure economic loss, discrimination, privacy violation, and intellectual-property infringement. Causation is often contested because AI outputs are probabilistic and value chains are long (foundation model provider to integrator to deployer to user).
| Feature | Detail |
|---|---|
| Published | 18 November 2024 (OJ) |
| Applicable from | 9 December 2026 |
| Covers | Software, AI, digital services, digital manufacturing files |
| Defect presumption | Possible when scientific/technical complexity creates disclosure asymmetry |
| Disclosure obligation | Defendants must produce "necessary and proportionate" evidence |
| Liable actors | Manufacturer, authorised representative, importer, fulfilment service |
| Theory | When It Applies |
|---|---|
| Product liability | Defective AI product causes harm |
| Negligence | Reasonable-care failure in design, deployment, or oversight |
| Breach of contract | AI fails contract specifications |
| Strict product liability | EU PLD 2024, US Restatement (Third) |
| Professional malpractice | Doctor, lawyer, engineer misuses AI |
| Vicarious liability | Employer for employee AI misuse |
| Statutory liability | GDPR, ADA, Title VII, consumer-protection |
Air Canada Chatbot (2024) — BC Civil Resolution Tribunal: airline liable for misinformation from AI chatbot.
Mata v. Avianca (S.D.N.Y. 2023) — Lawyers sanctioned USD 5,000 each after citing ChatGPT-hallucinated cases.
Park v. Kim (2024) — Second Circuit affirmed sanctions on lawyer citing AI-generated fake cases.
Uber ATG (2018) — Vehicle operator pleaded guilty to negligent homicide (2023); Uber settled civil claims.
Tesla Autopilot (ongoing) — Multiple US wrongful-death suits; jury verdicts split between plaintiff and defense.
Workday (N.D. Cal., pending) — Proposed class action for ADEA violations via AI hiring tools.
| Role | Core Duty |
|---|---|
| Foundation model provider | Safe defaults, accurate documentation, indemnities |
| Deployer (SaaS) | Implement required safeguards, document choices |
| Professional user | Human oversight, verify outputs before acting |
| Consumer user | Follow T&Cs; limited downstream liability |
| Provider | Customer IP Indemnity |
|---|---|
| Microsoft Copilot Copyright Commitment | Yes (for eligible customers) |
| Google Cloud Vertex AI | Yes (generative AI indemnity) |
| OpenAI Copyright Shield | Yes (ChatGPT Enterprise and API) |
| Adobe Firefly | Yes (IP indemnity) |
| AWS Bedrock | Yes (for titan and selected models) |
Q: Is the AI Liability Directive dead? Withdrawn from 2025 work programme; national tort law and the revised PLD continue to apply.
Q: Does EU PLD 2024 affect US companies? Yes if AI products are placed on the EU market.
Q: Can lawyers use AI without liability? Yes — but professional duties of competence and candour to the tribunal remain (ABA Model Rules 1.1, 3.3).
Q: Is AI output a product or service? Depends. Embedded AI in hardware is typically a product; pure SaaS is a service. PLD 2024 clarifies software inclusion.
Q: Do disclaimers eliminate liability? Rarely — consumer-protection statutes, professional duties, and PLD override unfair disclaimers.
Q: Who is liable for hallucinations? The deployer, typically, if a reasonable person would have verified the output.
Q: Is AI insurance available? Yes — multiple insurers offer dedicated products; premiums depend on governance maturity.
There is no single defendant for AI harm — there is a chain. Well-drafted contracts, documented oversight, and insurance close the gap.
Negotiate AI contracts confidently with Misar AI's AI Contracting Playbook.
Free newsletter
Join thousands of creators and builders. One email a week — practical AI tips, platform updates, and curated reads.
No spam · Unsubscribe anytime
Comments
Sign in to join the conversation
No comments yet. Be the first to share your thoughts!