Your Best Defense Against AI Errors in Contracts: Learn the most typical human mistakes

Gen-AI tools hallucinate, but that’s not their main problem when it comes to contracting. Rather, they reproduce errors typical of human contract-drafters (quite accurately, it seems). Worse, they repeat those errors without equivocation, confidently recommending bad terms and interpretations. The solution: learn the mistakes typical of human drafters, so you can catch AI mistakes.

Bell Curve vs. Pyramid

The reason, of course, is that human errors show up in the AI’s training data, in large numbers. In fact, human errors may present a bigger problem for contract-drafting AI than for AI tools in other fields, like news. Gen-AI seems to do a reasonable job summarizing the news – probably because most stories in its training data (or RAG data, etc.) have reasonably accurate content. And the inaccuracies likely balance out, with crazy left-wing stories canceling out crazy right-wing stories. The news, in other words, is likely to operate under a bell curve, with the good stuff mostly in the big middle area.

Contracts, on other hand, likely operate under a pyramid. Human errors appear in the bulk of online contracts, in my experience, and they’re at least common in online advice about contracts. That’s the base of the pyramid, and it’s far larger than the top: error-free contracts and sophisticated advice. But the AI can’t easily distinguish the base from the top, so it goes with the larger set: contracts and advice with errors.

Examples

Here are two examples of human-driven AI errors, both of which arose in testing for our courses:

Treating SaaS Like Services: One gen-AI tool for contracts recommended a professional services warranty – workmanlike services – for an agreement about SaaS. That suggests it took “service” in “software-as-a-service” to mean something like professional/human services. Human contract-drafters regularly make that mistake, so it probably outweighed better contracts and advice in the training data. In reality, “services” in SaaS doesn’t mean the same as in professional/human services, and arguably “hosted software” would be more accurate than “as-a-service.” (I know this confuses a lot of people because I’ve testified about it more than once as an expert witness.)
Calling for “Balanced” Indemnities: Another AI pointed out that a contract had indemnities protecting one party but not the other and said they should be balanced. That’s another typical human mistake, so again, it probably appeared in the training data. In reality, an indemnity in an IT contract just offers protection against third party lawsuits. So contracting parties only need indemnities against the suits likely to hit them. And if there are no such suits, or few, the party should spend its negotiation leverage on something more valuable, even if the other party receives indemnities.

Learning Typical Human Errors

The solution is to learn the errors most typical of human contract-drafters.

This is self-serving, since I provide exactly that sort of education (in my book, The Tech Contracts Handbook, and in our trainings). But it’s true. And I can offer another resource for free, right here.

Below is a list of the human errors we see most often. Obviously, it can’t cover the full range of those errors, but it puts a big dent in the problem.

Licenses to “Use” On-Premise Software: Rights granted or withheld should match the “monopoly” rights of copyright holders. And in the U.S. and many other jurisdictions, there is no monopoly over the right to use.
Vendor grants of “licenses” to SaaS, AIaaS, etc: Customers don’t reproduce SaaS, other cloud services, or most AI, so they don’t need copyright licenses. And even suggesting license rights can create problems for the vendor.
Payment terms that don’t account properly for revenue recognition
Vendors granting disputed payment terms: That’s just a get-out-of jail-free card for the customer, and you throw away leverage in a dispute.
Restrictions on prices distributors can charge: This creates antitrust issues and so, at a minimum, needs antitrust counsel’s review.
Customer failure to recognize that SLAs promise almost nothing
Customer failure to include warranties alongside SLAs
Including data in NDAs and confidentiality terms: Data should be covered by DPAs and the like. NDAs work for trade secrets.
Promises to comply with privacy policies or privacy laws: Neither requires your promise in a contract (except in a few cases re laws). And giving that promise just creates a risk that you’ll add breach of contract to the unpleasant consequences if you accidentally fail to comply (a likely event).
Focusing on ownership of data rather than control: You can’t own “your data” in a meaningful way, so you shouldn’t rely on it. You need terms controlling what the other party does with it.
Warrantors agreeing to “represent and warrant” … anything: You don’t want to represent, except in rare cases.
Treating indemnities like remedies for wrongdoing: Sadly, this is too complex to explain here.
Using liquidated damages as early termination fees: LDs are damages for breach. Early termination fees are not.
Defining “material” in termination for material breach: Defining it will usually screw up your common law rights (though you can add a cause supplementing the common law definition – e.g., “material breaches include, without limitation, ______”).
Customers requiring promises to give them no open source software: Good luck with that. OSS is almost certainly in the vendor’s product, and it almost certainly won’t do you any harm.
The services and indemnity errors above under Examples

Watch out for AI outputs that either miss these issues or that send you in the wrong direction.

And of course, please do join one of our trainings or pick up my book, to learn many more.

THIS ARTICLE IS NOT LEGAL ADVICE. IT IS GENERAL IN NATURE AND MAY NOT BE SUFFICIENT FOR A SPECIFIC CONTRACTUAL, TECHNOLOGICAL, OR LEGAL PROBLEM OR DISPUTE, AND IT IS NOT PROVIDED WITH ANY GUARANTEE, WARRANTY, OR REPRESENTATION. LEGAL SITUATIONS VARY, SO BEFORE ACTING ON ANY SUGGESTION IN THIS ARTICLE, YOU SHOULD CONSULT A QUALIFIED ATTORNEY REGARDING YOUR SPECIFIC MATTER OR NEED.

Share the Post: