How it works in a data room
Buyers ask the target for an inventory of AI systems in its products and internal tools, the sources and licenses of training data, contracts with model providers, evaluation results, incident logs and governance policies. Technical advisers may test models directly in a sandbox. Material such as model weights or proprietary datasets is rarely uploaded; it is described in documents and examined under controlled access. Findings link closely to intellectual property due diligence and privacy review.
Why it matters in a deal
AI can be a real source of value or a source of hidden liability. Training data used without rights can lead to claims, outside model terms may forbid the target’s actual use, and new rules such as the EU AI Act can bring compliance duties. Valuations that rely on AI claims should be backed by evidence. This is distinct from using AI tools to speed up diligence itself.
Example
A buyer considering an AI-driven claims processing startup learns that much of its training data came from a customer whose contract forbids reuse. The buyer requires written consent from that customer before closing and a specific indemnity for past use. The technology and software guide covers technology targets.