Latest / B2B SaaS Talks with Fexingo: Enterprise Software, Sales Cycles, and Procurement / Why Enterprise RFPs Now Probe Vendor AI Copyright Indemnification
Transcript
- Lucas: So there's a new clause showing up in enterprise RFPs that's making a lot of SaaS vendors sweat, and it's not about data residency or SOC 2 reports. It's about AI copyright indemnification. Luna: That sounds like the legal team finally got their hands on the AI strategy. What's driving it? Lucas: Right. We've seen the usual asks — training data provenance, model rollback, even climate transition plans. But indemnification is different. It's not about what the AI does, it's about who pays when the AI's output infringe someone's copyright. Luna: So buyers are saying, 'If your model generates something that gets us sued, you're on the hook.' That's a big ask. Lucas: Exactly. And it's not just the big players. I've seen a mid-market manufacturing company's RFP include a 40-page indemnification appendix. Forty pages, Luna. That's more than their entire security questionnaire. Luna: Wow. So how do vendors even respond to that? I mean, standard indemnification is usually capped at the contract value, but AI copyright claims can blow past that in a single lawsuit. Lucas: That's the crux. Traditionally, indemnification is capped at the fees paid over the last twelve months. But with AI, the exposure is potentially unlimited. So vendors are getting creative with carve-outs and caps. Luna: What kind of carve-outs? Lucas: The most common one is around model provenance. Vendors are saying, 'We only indemnify against claims that arise from our training data, not from user prompts.' That shifts some risk back to the buyer. Luna: But buyers are pushing back on that. If they're using the product as intended, why should they bear the risk? Lucas: Right. And that's where the negotiation gets interesting. Some vendors are offering a tiered approach — full indemnification for standard use cases, but only if the buyer agrees to audit the model's outputs periodically. Luna: An output audit? That's a new one. How does that even work? Lucas: Basically, the vendor provides a tool that flags potentially infringing content and reports back to the buyer. It's like a content filter, but with a legal purpose. If the buyer runs it and still uses the output, the vendor's liability is reduced. Luna: So they're turning the indemnification into a shared responsibility model. That's clever. Lucas: It's also a way to differentiate in a crowded market. We're seeing vendors advertise 'full AI IP indemnification' as a feature, like it's a security badge. Luna: But there's a catch, right? Because no one can truly guarantee that their AI won't reproduce copyrighted material. Lucas: Exactly. That's why some legal teams are laughing at these clauses. The vendor is promising something they can't control. In the AI world, the model is a black box, even to its creators. Luna: So what's the realistic middle ground? I've heard some buyers are asking for a separate AI indemnification fund, where the vendor sets aside a pool of money to cover potential claims. Lucas: Yeah, I've seen that too. Usually it's a percentage of the contract value, like five percent, held in escrow. If no claims within three years, it releases back to the vendor. Luna: That's a nice compromise. It gives the buyer some assurance without requiring the vendor to assume unlimited liability. Lucas: But some vendors are balking at that, because it ties up cash and adds administrative overhead. And a lot of startups just don't have the balance sheet to support it. Luna: So is this becoming a differentiator? I mean, the vendors who can offer robust indemnification might win deals, even if their product isn't the best. Lucas: Definitely. We saw a similar pattern with SOC 2 reports a few years ago. It started as a checkbox, then became a differentiator, and now it's table stakes. I think AI indemnification is on the same trajectory. Luna: But unlike SOC 2, this is a real financial risk. A single copyright suit could bankrupt a small vendor. Lucas: That's why we're seeing vendors get creative with insurance. There's a new class of 'AI liability insurance' that's starting to emerge. Some vendors are buying it just to be able to say 'we're covered' in their RFP responses. Luna: And how do buyers react to that? Is it enough? Lucas: It's a start, but the insurance policies have so many exclusions that some lawyers joke they're not worth the paper they're printed on. Still, it's better than nothing. Luna: So if I'm a SaaS vendor, what's my move? How do I navigate this without signing away my company? Lucas: First, don't ignore it. We're seeing RFPs that explicitly require a response to the AI indemnification clause, and if you leave it blank, you're disqualified. Luna: That's harsh. But it makes sense — they want to see that you've at least thought about it. Lucas: Second, know your model. You need to understand your training data and your model's tendency to reproduce verbatim text. There are tools that can test for that. Luna: So being transparent about your model's limitations works in your favor? Lucas: It does. Buyers respect honesty. If you can say, 'We've tested our model and here's the infringement rate,' that's more persuasive than a blanket promise. Luna: And then you can negotiate a cap that's proportionate to the actual risk. Lucas: Exactly. We're seeing some vendors offer multi-tiered indemnification — a lower cap for standard use, and a higher cap if the buyer pays for a premium support package. That aligns the risk with the revenue. Luna: So it's not just about the legal language; it's about your whole go to market strategy. Lucas: Right. And on the buyer side, we're seeing procurement teams start to understand that they can't ask for the moon. They need to balance their risk with the vendor's ability to deliver. Luna: It's a learning process for everyone. In a few years, we'll probably have standard language for this, like we do for data processing agreements. Lucas: That's the hope. Until then, it's a wild west out there. Every RFP is a negotiation, and the vendor who can articulate their position clearly is the one who wins. Luna: And honestly, if this episode helped you think through your own RFP strategy, that's the kind of practical insight we aim for. If today's conversation was worth a coffee to you, the link is buy me a coffee dot com slash fexingo. Lucas: Thanks for saying that, Luna. It does keep the lights on around here. And speaking of practical insights, let's talk about what happens when the clause actually gets triggered. Luna: Oh, the enforcement side. That's a whole other beast. Lucas: Right. We've seen a handful of cases where a buyer invoked the indemnification, and the vendor had to step in. One case involved a marketing firm that used an AI tool to generate ad copy, and it copied a slogan from a competitor. The vendor settled for a mid-six-figure amount. Luna: Mid-six-figures. That's not trivial. And that was just one claim. Lucas: Now imagine a hospital using an AI diagnostic tool that produces an infringing image in a patient's record. The exposure is much larger. So the stakes are real. Luna: So what's the bottom line for our listeners? If they're a vendor, should they be scared? Lucas: Not scared, but prepared. The ones who are ahead of this are the ones who'll win the deals. They have a clear policy, they've stress-tested their model, and they're ready to negotiate. The ones who ignore it will find themselves on the losing end of the RFP process. Luna: And that's a risk no vendor can afford. Lucas: Exactly. So, as we head into the fall procurement cycle, make sure your AI indemnification answer isn't just a boilerplate paragraph. Treat it like the strategic asset it's becoming.