Liability Vacuum

Software manufacturers have never faced meaningful product liability. Contracts disclaim the warranties and cap the damages, so the cost of a defect stays with the buyer.

Since AI is software, that gap applies directly to AI. Agents widen it. When an agent causes an incident, liability runs through the model provider, the application developer, the deploying organization, and the operator, and no existing framework cleanly assigns it. This is the manufacturer/operator problem, amplified (see C-15).

Discussion prompts

• Where does accountability live in your organization when an AI system causes harm?
• How would you trace responsibility through a chain of agents built by different vendors when something goes wrong?
• What would meaningful AI product liability look like for software you build or buy?