The Short Pour - Workday & Meta Lawsuits

[00:00:00] Welcome to The Short Pour, Whinos! Today we are discussing AI antics, the Workday and Meta Lawsuits. Julie and I have spoken about AI in the past, check out episode 9 from this season, and we've discussed how AI is inherently biased because the people who are training it are also biased.

[00:00:28] Now, we do not think that AI is inherently bad, but it is only as good as the people who are training it or the people who are feeding it prompts. And because all of us are inherently biased, you see the problem. Two big companies, Workday and Meta, have recently learned this lesson the hard way. So, let's dive in.

[00:00:54] Rich, can you give us an overview of Workday's lawsuit? I sure can! Okay, so I want to make sure I get this right. If you see me occasionally glancing at my notes just to understand it's because I don't want to misquote. I want to make sure because this is super important.

[00:01:11] Rich, can you remember this lawsuit in the case. This lawsuit was filed by a Derek Mobley. He was a Black applicant, he was over the age of 40 and he has a disability. Now what he alleges is that he applied for more than 100 jobs for different employers using Workday's recruiting platform. Okay?

[00:01:36] Now, the problem is he was repeatedly rejected, often within minutes without even having interviews. So what he's claiming is that Workday's AI-driven screening, and they also have recommendation tools, they disproportionately disadvantage applicants based on their protected characteristics. We're looking at age, race, and of course, disability.

[00:02:06] So why this is important is because a federal judge actually allowed these key discrimination claims to move forward, finding that the allegations were actually plausible enough to proceed under theories such as disparate impact discrimination. Remember, when we're talking about disparate impact, we are talking about an impact on one or more of your populations

[00:02:34] that impact then affects them based on their protected characteristics at a higher rate. So that's what we're talking about with disparate impact. Now, the plaintiff does argue that the AI tools themselves, they score, they rank, they sort, and they screen candidates.

[00:02:56] And they're basically doing that in a way that disadvantages older applicants, black applicants, those who have disabilities, and also women. Now, Workday actually does deny any sort of wrongdoing and says that its technology evaluates job qualifications rather than those protected characteristics, such as, again, race, age, and disability.

[00:03:24] So it'll be very interesting to see what happens with this one. Yeah, so now the next one is the meta lawsuit, which does not deal with hiring and recruiting, but deals with layoffs. So this one is pretty interesting as well, and I am also going to consult some notes, so you might see me occasionally looking down.

[00:03:46] But in July of 2026, 26 current and former meta employees sued the company, alleging that AI systems were used during a layoff, and that those systems disproportionately selected employees who were on either protected medical leave, so things like Family and Medical Leave Act, family, parental, or disability leave, or that had an accommodation under the Americans with Disabilities Act.

[00:04:16] So those employees were part of a layoff that affected roughly 10% of Meta's workforce or about 8,000 employees. So we are not talking small potatoes here. So what discrimination specifically is alleged? The lawsuit alleges that Meta relied on a collection of AI-assisted evaluation systems and productivity metrics

[00:04:39] that included performance scores, productivity measurements, activity monitoring data, AI usage metrics, which is like AI native or token consumption measures. I don't quite frankly understand what any of that means, but I'm sure there are people out there smarter than me who do, and algorithmic ranking systems.

[00:05:05] So, again, according to the plaintiffs, employees on medical leave, pregnancy leave, parental leave, or who had disability accommodations naturally generated fewer of those productivity measurements, and as a result, they were ranked lower by this AI system and made more likely to be selected for layoffs.

[00:05:27] And I have seen this in other non-AI related cases where certain performance measurements are created, and then there isn't any way to take into account for people who physically aren't present to be able to meet or match those accommodations.

[00:05:49] So there was a case out of, I don't remember which district it was, it was a federal case. The plaintiff's name was Schwartz, and he was a maintenance guy for a particular manufacturing company, and he was out on FMLA caring for both of his parents who each had their own serious health condition. So the poor man had a lot going on.

[00:06:14] While he was out on FMLA caring for his parents, his supervisor, first of all, sent some emails to HR saying, how do I get rid of this guy? Basically, that was the gist. But then also, so how he decided he was going to do that is he created these new performance metrics where each maintenance individual had to complete a certain number of work orders every single month.

[00:06:39] And if they didn't, then they would get ranked lower and they would be at risk for discipline or termination. Well, because he was off on FMLA, he obviously could not get the same amount of work orders done as somebody who was there full time. He wasn't given any option to prorate the amount of work orders or anything like that. It was just like, you didn't make it, you're gone. And so he sued and he won.

[00:07:04] And this is kind of the same thing that those AI things are doing in what is being alleged in the Meta lawsuit. So like Workday, Meta has said, we didn't do anything wrong. Those decisions were made by people, not AI, et cetera, et cetera. So there was a recent update. The judge refused to block the layoffs.

[00:07:30] The plaintiffs had asked for a temporary restraining order to block their layoffs. The judge said that wasn't necessary because they could not show harm that could later be addressed. And while that is true, they can get damages for, you know, if they are found to prevail on their claims, they could get damages for whatever money they have lost. The simple fact of the matter is they are currently without jobs, which means they could be in precarious positions.

[00:07:59] So I'm not sure that the judge got that 100% right, but we shall see how that plays out. So all that being said, Trish, taking away from this, what are your tips for employers? I think my number one thing would be this. Look, I understand AI can be so incredibly helpful, but you have to have some oversight. Julie and I did talk about that during our AI episode. You must have some oversight.

[00:08:26] So if you are going to use AI to assist in some of these decisions that you're going to make, make sure that you set the metrics properly. Make sure that you define everything and think about where can this bias creep in and make sure that once it is, okay, we have chosen these things. Why are we choosing these particular metrics? And then again, a triple check to see, okay, now did this have a disproportionate impact

[00:08:56] on certain members of our population? Yeah, I think very similarly, I would seek out people in some of those other populations. So for example, if you do have employees who do have disabilities, have been on medical leave, have been on, find a couple of those individuals that you trust and have them run through the algorithm and see if it works or doesn't work for them.

[00:09:21] So one of the biggest things I've heard with respect to disabilities is that oftentimes when you have these prompts, like even in a hiring situation, they won't give, they only give you a limited period of time to answer one question to the next question. And if you take too much time, sometimes it'll boot you right out. And just eject you from the system. And if you have a disability, and for some reason, maybe it's a processing disability, maybe

[00:09:49] it's just that whatever the case may be, however that looks, if you are not allowing for them to ask for more time to ask for an accommodation to ask for any sort of reprieve from whatever that algorithm says or whatever that, you know, prompt is, then that is potentially problematic.

[00:10:14] So you really want to think about run through a checklist of, okay, let's go through all our protected classes. Let's go through all of the potential problems that could arise from this. Have a bunch of employees tested out and tell you what they found worked and what didn't work or where they got the ics or where they got that, you know, like whatever the case may be. Test it. Test it. Don't just assume it's going to be flawless because what we're learning is AI is only as

[00:10:42] good as the people who create the prompts. And we are all flawed as we've talked about time and time again. So with that, hope we gave you some food for thought. Don't forget to subscribe wherever you get your podcasts. Follow us on Facebook. Follow us on where else are we? Instagram and LinkedIn, YouTube, all the places, all the places till next time. Cheers.