If you’re going to farm out your company documents to AI, at least make sure the AI isn’t confessing to any labor law violations.
Matt Bruenig flagged a weird sign-of-the-times NLRB matter out of Texas. Daniela Melendez worked for an aftermarket auto parts company called Autofit as an administrative assistant at $22 an hour. Two or three days in, another employee asked what Melendez was making. Melendez told her, and the other employee printed out Melendez’s job posting, marching into a boss’s office and asked to be paid the same. Management told other employee that she didn’t qualify.
Sixteen days later, Melendez was fired.
When Melendez applied for unemployment, the company — via Chief Logistics Officer Elham Emadi’s administrative assistant Esperanza Guevara — had to explain itself to the Texas Workforce Commission in writing. This is where things get screwy:
The employee was terminated due to a number of performance issues, which included serious breaches of confidentiality and difficulties in teamwork. As part of her role, she was explicitly instructed to keep sensitive information confidential, yet she shared pay details with colleagues.
According to the decision, after Melendez learned about this rationale, she asked whether talking about pay was against the law. The answer is, of course, no. But firing someone for talking about pay very much is against the law.
Autofit had to explain why it admitted to breaking the law. Its first attempt came in a sworn statement taken during the Board’s investigation, where the company argued that when it wrote “she shared pay details with colleagues,” it didn’t mean talking about pay, but “taking information or sales commissions of others and sharing it or using it for herself.”
Since that’s not how words work, this didn’t fly.
Autofit’s own lawyer teed up attempt two at trial: you see, the company hadn’t admitted anything because ChatGPT wrote it! The author of the statement testified that she used the free version because she wanted a more professional sounding response, and then copied and pasted the output directly into the form. She couldn’t recall whether she read it thoroughly, and figured ADP would fix any errors. She had not mentioned any of this to anyone until witness preparation.
Hallucinations are wild… but they aren’t that wild:
Guevara would have us believe that ChatGPT made up the language that sensitive information includes pay. For example, why didn’t ChatGPT say another type of sensitive information, like trade secrets. I therefore do not credit Guevara’s explanation that ChatGPT added pay without Guevara’s help.
That’s Judge Sharon Steckler not buying the idea that ChatGPT just decided to make up a reason to fire Melendez without any prompting. An illegal reason to fire her at that.
Even if ChatGPT spit out language about pay, Guevara could have reviewed it, found it inaccurate, and removed the part about pay.
With the company in the bind between admitting that it filed a termination explanation with the state including fake justifications or conceding that the explanation included the genuine admission that the company fired her illegally, the judge credited the latter explanation.
The company also lodged 16 affirmative defenses, including a Seventh Amendment jury trial demand and the now-standard Article II claim that the Board itself is unconstitutional because presidents are god emperors with unchecked power over the executive branch as long as they’re a Republican. Neither carried the day.
The remedy is reinstatement, backpay, pecuniary harms, file expungement, and a notice on the wall for 60 days.
There is one more detail worth sitting with. Autofit’s witnesses testified that the company demands employees be trustworthy above all, and the administrative assistant job was described repeatedly at hearing as a sensitive position. Outsourcing a state agency filing to a free chatbot was apparently not a breach of sensitivity, but talking about your own paycheck was.
Garbage in, garbage out.
(Check out the opinion here…)
Joe Patrice is a senior editor at Above the Law and co-host of Thinking Like A Lawyer. Feel free to email any tips, questions, or comments. Follow him on Twitter or Bluesky if you’re interested in law, politics, and a healthy dose of college sports news.