What is AI liability?

At a glance

• If your AI gives bad advice or makes a harmful decision, your business usually owns the cost^[4].
• Courts won’t accept “the AI did it” — a tribunal held Air Canada liable for its chatbot’s wrong advice^[1].
• The EU is shifting to no-fault liability: AI now counts as a “product,” so harm can cost you even without proven negligence^[2].
• Liability rarely passes to the AI vendor unless your contract says so^[5].

Who pays

Claims usually run under product liability, negligence, or misrepresentation. The candidates are the AI’s maker, the business deploying it, and sometimes the user — but courts most often point at the company in front of the customer^[1].

The law is tightening

The EU’s revised Product Liability Directive (2024/2853) treats software and AI as products, so a harmed person need only show a defect, not your carelessness; member states must adopt it by December 9, 2026^[2]. A separate AI Liability Directive was proposed but withdrawn in 2025^[3].

What to do

Treat AI outputs as your own statements. Keep humans reviewing high-stakes decisions, document oversight, add clear disclaimers, check who absorbs liability in vendor contracts, and confirm insurance covers AI errors.

Bottom line

If you deploy AI, assume you own what it does.