Typically, when we describe a company as “conservative,” we’re talking about a company refusing to use the cloud because they’re imaging all the servers falling into the sea. So it’s a little hard to accept the adjective being applied to a company trying to turn over its contract negotiating to fast-paced hallucination machines.
Spellbook CEO Scott Stevenson shared a highlight from this week’s sales call:
This is a little shocking. Than again, an industry that’s based upon gouging people with terminal illnesses enters with a fairly dim view of humanity so maybe this tracks. UnitedHealth faces a proposed class action alleging it used an algorithm called nH Predict to cut off post-acute care for Medicare Advantage patients, with the complaint claiming more than 90 percent of those denials were overturned on appeal.
To be clear, Spellbook isn’t selling a solution this radical. Its newly released “Autonomous Contract Management” product reviews counterparty redlines the moment they come back and checks them against a playbook — a genuinely useful AI application — but then it marks routine deals “ready to approve.” And by “ready to approve,” they mean have a human lawyer practice real judgment. This “conservative” client apparently longs for an option that skips the approval button.
The tweet doesn’t clarify the role of this anonymous inquisitor. One hopes it wasn’t the general counsel, openly pining for an opportunity to engage in a little unauthorized practice of law. In-house counsel get a bad rap as the “Department of No,” but this course correction should stop short of “Department of No Worries.”
Later in the comments, Stevenson clarified that the client was only talking about “NDAs and simple MSAs,” which are admittedly fairly boilerplate. On the other hand, streamlining NDAs sounds like the beginning of a story that ends with Harvey Weinstein. Business confidentiality agreements aren’t the same as buying a contractual coverup, of course, but one of the undersold lessons of the Weinstein debacle is that these agreements cause damage in large part because everyone treats them as boilerplate that nobody bothers to read. The stakes may be much lower in these agreements, but the vector to causing damage is the same.
And there’s the added risk of opening the company up to prompt injection. Morgan Lewis recently warned that hidden text buried in a document can steer an AI away from its user’s instructions and flagged contracts as a key exposure point. The firm did not think any current product can be assumed to reliably catch the problem. That may or may not be true, BUT it is true that every time a product adjusts to fix a vulnerability, someone else is going to invent a new avenue to get behind that wall.
Which is how AI tools that speed up the process, while maintaining human involvement, should improve outcomes. A human reviewing the 400th NDA of the afternoon is mostly skimming these things anyway. Using technology to give the human an easier and more informed starting point on each agreement almost certainly improves the chances that nothing gets missed. And, of course, the human can see when the 399th agreement looks weird.
But this tweet is still a little scary. There are clients out there — with money — who want to throw caution to the wind. Well, correction: that’s scary for transactional lawyers.
For litigators and bankruptcy attorneys this could be a real boon!
