Short Summary:
When AI Transparency Becomes Control:

A vendor sends out its AI addendum and receives a redline requiring it to disclose the identity and version of each third-party foundational model used in the service.

The clause also prevents the vendor from replacing or materially modifying the model without the customer’s prior written consent. This appears to be a transparency obligation. In practice, it gives the customer a potential veto right over one of the core components of the vendor’s product architecture.

That level of control has rarely existed in traditional software and SaaS transactions. Customers generally assess the finished product while relying on the vendor to select and manage its development tools, infrastructure providers, third-party software and other elements of its technology stack.

Customers are no longer treating the LLM as an invisible component within the vendor’s product. Many have evaluated the leading model providers and created internal lists of approved and prohibited models.

The first question in the negotiation is therefore whether the customer is asking for reasonable transparency or control over the vendor’s product architecture.

The video explains why customers are making these requests and why the underlying model has become a material procurement issue.