New Tools, New Liabilities

AI is not a software upgrade. It is a new category of business tool with a new category of legal exposure. Traditional contracts were drafted for deterministic systems — software that does exactly what the specification says, every time. AI does not work that way. It generates outputs based on probabilistic inference, which means the same input can produce different outputs, and nobody — not the developer, not the deployer, not the client — can guarantee what it will say next.

This matters legally because most existing contract language assumes predictable behavior. Warranty clauses promise the software will perform "as described in the documentation." Indemnification clauses allocate risk based on whose product caused the harm. Limitation of liability provisions cap exposure based on what could reasonably go wrong. AI breaks every one of these assumptions. The documentation cannot fully describe probabilistic outputs. Determining whose product "caused" an AI error is genuinely ambiguous. And what can "reasonably go wrong" with a system that generates novel content is a question no court has definitively answered.

I am not telling you this to create anxiety. I am telling you this because the organizations that address these gaps proactively will spend a fraction of what the organizations that discover them reactively will spend. A contract review costs hours. A contract dispute costs months. Read before you sign. Always.