Nate Kostelnik’s Post

2 days ago I attended an eye-opening session about AI vendor agreements and how to redline them. Here are my 3 biggest takeaways (along with a bonus takeaway if you're worried about AI replacing lawyers and legal pros)... ✅ The size of the AI vendor's model may dictate the warranties they're willing to give. You likely won't get a warranty of non-infringement of outputs with larger models, but you might with smaller models. This makes sense: the less stuff you have in your model, the more you should know about it. ✅ IP indemnity for AI outputs seems to be the growing trend. Of course, there are limit. For example, if you put in infringing content or turn off system guardrails, no indemnity. But otherwise, if you use the tool as intended and the output is infringing, IP indemnity should be available, at least from the major players. ✅ Customers want assurances that vendors won't train their AI models on the customers' data. If you're a customer, how do you get this assurance? Standard confidentiality clauses probably aren't sufficient. Allowing the vendor to use the customer's data only "to provide the Service" or "to improve the Service" is squishy and gives vendors an argument that they can still train on the customers' data. The best option? State the restriction clearly so there's no confusion: "Vendor shall not train..." The superstar⭐ presenters David Tollen, Laura Jeffords Greenberg, and Nada Alnajafi were fantastic as always! And David offered not only my favorite takeaway, but also an eye-opening perspective that changes the way I think about AI going forward: 👇 AI prompting is a skill. And a big part of that skill is CONTEXT. If you prompt AI and don't give it enough context, you'll likely be disappointed in the output. You can improve the output by giving AI more CONTEXT. And guess who is pretty good at providing context? Contract drafters! Contract drafting is all about providing context! No, not the type of context that you see in those old, stodgy Whereas clauses ("Whereas, Buyer wishes to buy, and Seller wishes to sell..."). But the context we establish through clear, thorough drafting about the parties' relationship and the transaction: • Here's the parties' mutual understanding of what the deal is about. • Here are the risks the parties are concerned about. • Here are the circumstances when this particular provision will apply. • Here's how this provision relates to this other provision. We provide a ton of context through drafting! There's a lot of hullabaloo about AI replacing lawyers and legal pros, but we should recognize that the same skills that help us draft contracts can also help us make us really good at AI prompting. 💡 Creating a good prompt is not that different from drafting a precise, unambiguous clause. In both cases, we have to provide CONTEXT to ensure that we're on the same page as the counterparty or AI. Thanks to Contract Nerds for the great webinar!