Crossing the AI Minefield
You need to start reading your contracts, and your writing, very carefully.
Nota bene: As I already have a base of paid subscribers, I really do owe them paid content, which will focus on my day job, the craft and business of storytelling. How to make something good, sell it, and maintain your sanity. Every paid subscriber means I can keep procrastinating on my real job; getting my ass kicked making and selling things.
A writer friend is in the process of optioning a tv project to a particularly ruthless company. As writers do, we talk regularly about our deals and rant to each other about how crazy some of the terms we get offered are.
Put briefly: you have to be really, really careful with any AI use whatsoever. Your AI hygiene must be meticulous, because even suspected AI use is already tanking several major deals. Not even proven AI use, just the mere suspicion.
And my god, who at this point has not felt that suspicion? Where there’s smoke, it seems, there’s usually chatbots.
That smoke now seems to be costing people small fortunes.
So let’s get to the contract. You, your reps and your lawyers are going to have to read these things very closely, and maybe even get into a fight.
The issue really revolves around two clauses.
A Brief History of the Two Major AI Blowups This Year
The first is the main clause covering and disclosing the writer’s use of AI. This makes a lot of sense after the aforementioned cataclysmic meltdowns of two high profile publishing projects, detonating millions of dollars overnight.
The first novel in this new trend came in March, with Mia Ballard’s initially self-published horror novel Shy Girl, about a young woman held hostage by a sugar daddy who forces her to live as his pet, which the NYTimes reported Hachette as having pulled over AI claims.
The most recent one was for a book by Jerry Adedayo Falade called Call Me, I’ll Hide the Body, which followed a character not unlike the Wolf from Pulp Fiction who, you guessed it, hides bodies. It was supposed to sell for $2.4m and sparked a 14 way bidding war. It will likely now sell for nothing.
Screenrant wrote that:
An editor also raised concerns about possible AI use, leading to a phone call where the agent alleges that [Call Me’s] author admitted to using AI for light research. Afterward, the agents decided to pull the book, saying they couldn’t substantiate whether AI was used in the actual writing.
This is especially terrifying because there has yet to be conclusive evidence presented that the novel contained AI writing.
The problems arose, again, from the mere suspicion of generative AI use. The agents did not run it through an AI detector, but Publishers Lunch did. It returned that it was 99% AI.
This of course creates a serious issue for both copyright and chain of title.
Screenrant again:
AI muddies the waters because the person who prompted AI to create the manuscript doesn’t own the copyright. While laws vary from country to country, RAND confirms that the US, European Union, Japan, and South Korea all uphold the principle that human authorship is required for copyright. Writing created by AI is not protected, and putting a prompt into a machine doesn’t constitute human authorship.
Falade has vehemently denied the claims that he used AI, but the deal seems dead nonetheless.
The Deadline reporter is blunt:
From what I’m hearing, this will be an important teaching moment for writers with aspirations to become serious bankable authors. The lesson: stay the hell away from AI.
Strict AI Clauses
In my friend’s aforementioned contract, it’s expressly stipulated that the writer not use generative AI in the creation of the project.
It also requires that he inform the company of any use in the making of it. Any additional use of AI content would require their express, written consent. Even in the event of their consent, they require him to revise those results into his own writing.
Finally, he’s required to notify them of all AI use, at the latest upon delivery of the work, including which parts were created with AI, which applications were used, and a log of prompts entered.
This is perhaps a little excessive but fair. If companies can’t truly own AI writing in the same sense they can own and exploit human writing, they have to shift the legal liability for its use onto the author. It’s the only way to know without hacking into their computer.
So how do they detect it? Can they do so accurately?
And what’s that horrible second clause?
