AI Governance for HR: Avoiding a $365M Litigation Risk

So the question I wanna ask while we're waiting to start, we'll give it one more minute and then we'll start. Tell me your biggest challenge around AI in your workplace or AI governance in general. Just put that into the chat and just tell me what your biggest challenge is right now. And then please add what you want from this session. Okay, this is a high-level brief, but I want to know what you need so that I can customize this for you as well, um, so that I'm answering your questions. I'm leaving time for, for Q&A, um, and as we go through the process, what do you need me to cover that would make your job easier and better? So if you want to put that into the chat, that'll be great. And then we'll get started. So today's program is AI, HR, and AI governance. What happens when AI joins HR litigation? And we're gonna go through a bunch of cases that have occurred in the last few weeks, and we're gonna talk through the new shape of HR, and I'm calling it HR 2030, and who you must become to manage AI in the workplace. So we're going to really deep dive into this. For those of you who have not attended one of my programs before or who are watching the live or the replay, my name is Margaret Spence. I'm the founder of the Inclusion Learning Lab. I'm also running a brand new Substack called AI Governance for HR. We, we started a month ago. We have over 100 subscribers and we are writing every single day about how AI is governed in the workplace. For those of you joining us globally, we have a few people from the European Union, from Southeast Asia that signed up for this program. Your whole premise around AI governance is going to be very different in the coming months and coming weeks, and we are going to not make this program only US-centered. I'm going to touch on what is going on outside of the US and some new happenings that's, um, going on in the U.S. So let's talk and let's start. So many of you know that I've done diversity, equity, and inclusion in HR. I'm at a 40-year career and I keep, you know, pinching myself that I've been doing this work for 40 years. Um, it feels like I started when I was 2. But let me give you the reason why I pivoted and brought AI governance to the forefront. A lot of people are writing things about prompting, about using AI. My background is in risk and insurance management. I started my career as a risk manager managing large loss litigation for organizations globally. In my 10-year career, I have probably handled close to $1 billion in losses for organization. The largest risk case that I managed was a $400 million building risk where the builder built the building and totally defective building. And I was able to manage that case and bring it to a settlement for my client customers. So I come to HR not from an HR lens, I come to HR with a very solid risk and governance experience. For many years, I sat on the risk management committee for SHRM. I also wrote the questions that are in the SPHR HRCI certifications for risk management, governance, and HR. Managing workplace risk. So this is not new for me. A lot of people are, oh my gosh, I'm an AI governance person. I have been doing governance for the last 30 years of my career. And so I understand risk. I understand what that is. Let me give you one step further. I was on the pandemic response program that was put together long before we had the pandemic. It was actually put together by George W. Bush, and he appointed 400 of us across the U.S. to manage pandemic risk and the risk associated with pandemics. We weren't necessarily called when the pandemic happened, but I have been doing risk management for years. So I— and I've been doing HR for about 27 years. So the combo of that gives me a different lens about AI. I'm writing a new book. The new book is called When AI Breaks the Law. I think the book is on point. It is the first book that is being written specifically about AI governance for HR. And many of you have AI governance committees and it is run by IT. I am saying to you that HR needs its own AI governance, and that is what we're going to really discuss today. So let's get started in the meat and potatoes of this program. We have a $365 million wake-up call, and 2026 is the year HR gets sued because of AI. Now, many of you are aware of the Mobley v. Workday case. A good portion of my book is written about Mobley, but there are several other litigated, um, cases that we need to know. I wanna preface this by letting you all know that I am not an attorney. I am a risk manager who does HR, and as a risk manager, I look at litigation every single day. Let me give you my cadence around my world and, and how I happen to be where I am. Before there was ChatGPT and all these other things, I had a Google alert set every morning that tells me about HR litigation. That alert comes into my inbox at 6:00 AM every morning, and I look for workplace litigation that is occurring based on workplace injury, meaning workers' compensation, risk management, desperate impact, pregnancy discrimination, age discrimination, the long list of various discriminations. So I get this email every single morning, and I have since Google, I started using Google, and Google had this alert system for years. Last year I got a Bing, on Mobley v. Workday. It was certified as class action. And when I saw that, I thought, class action involving AI discrimination? Wait a minute, this is new. This is something different. So I read the entire case. I went and looked at the briefings. I pulled everything I could read from FindLaw, and I subscribe to FindLaw, which is where most attorneys go to get their case documentation. So I'm looking at the entire case. I'm looking at everything. And because I have been a risk manager and an adjuster, I have access to a lot of legal briefings that the normal person might not have access to. So I read it all. I reached out to a bunch of attorneys that I knew and I asked them about this, and everybody was in this really foggy place. So I started submitting to HR conferences, we need to talk about AI litigation involving HR. And I could tell you I submitted to 40 different conferences. And every single one turned me down. It was like, we are in, we don't want that, we want prompting. Well, prompting isn't going to keep us from being sued. So the Mobley case, Mobley v. Workday, has a $365 million exposure for Workday. There are 1.1 billion applicants who were affected by that case. And that is the exposure that Workday has. Workday has, in my opinion, done a very good job of protecting its customers from this litigation. I don't know that that protection is going to stand forever. So let's unpack all of the cases that are significant to us in HR. So imagine for a second it's 9:00 AM. And an algorithmic discrimination lawsuit has come into your office, to your company overnight. You've gotten this litigation. Your CFO, CEO, your legal, internal legal counsel shows up in your office as the HR leader and says, we have a discrimination claim. Our AI tool rejected 100,000 qualified candidates. And someone is alleging that they were discriminated against because they were over age 40. And you're sitting there going, okay, I get it. I'm not sure what I'm supposed to do right now, but your name is listed as the accountable party in that litigation, meaning the attorney for the plaintiff has picked you out as the CHRO or the HR leader and put your name on the litigation. So the question that we have to be able to answer: are you prepared to answer for the black box? Every AI that is being used in HR right now today is inside of a black box, and we as HR leaders have no idea what is actually going on inside of the black box. So let's unpack it. The first case is Mobley v. Workday. It basically, um, Derek Mobley, 40-year-old software engineer, African American, well experienced, applied for 100 jobs using the Workday platform. He received rejections at all hours of the day and night, generally within minutes of applying. When he got his 100th rejection at 1:50 AM, he knew that no human being was reviewing his application. The resume screening process was screening him out for some reason. And being the software engineer that he is, he decided to unpack the software. I'm not sure how he unpacked the software, but he did., and in unpacking the software, he felt that there was discrimination. So he filed this as Desperate Impact. You're basically picking me out and you're biased against me because of my age. The case went for a year of litigation before, um, a judge said, yes, you're right, um, we're going to certify this for class action. And then Workday had to produce how many applicants were actually screened by the AI tool. There's a couple components to this case. There is the AI screening tool, and then there was the psychometric evaluation done during the application process. We're not sure which one of the two things caused the discrimination, but he's alleging that the entire process cause the discrimination. Now, my husband works for one of the big box stores, and when he was applying for his job, I was helping him with his application, and they asked him psychometric tests. They asked him 40 questions that I would consider to be psychometric testing to figure out who he was. So every applicant in some form is going through some kind of screening The question is, do we know what decisions the AI is making after it does the screening? I had a huge argument with someone, with a recruiter on LinkedIn a couple weeks ago when I brought up the Mobley case, and she says, oh, we don't do that. No, no AI is screening our applications. Okay, you might be the exception. That is not the rule. A lot of you are using applicant tracking systems that have AI components built into them that screen resumes. Let me age and date myself. I can remember when, uh, Treo, I think it was called, way back when, 20 years ago, we would, um, keyword in. I would be looking for someone and I would put in the keywords in the back end of what I wanted on that resume, and it would sort and give me resumes with those keywords., and it made the process easier. So anybody that thinks, oh, AI is not in our process, it very well is there. The second case was EEOC v. i-Tutor Group. Um, they were charged with automated ageism. An AI program instantly rejected female applicants over age 55. And male applicants over age 60. It resulted in a federal settlement. Um, the algorithm was making human decisions that violated federal discrimination, ADEA, and delegating the decision to code. Okay, so this is case number 2, right? And we think, oh, it's not happening. This one resulted in a settlement. The next one was CVS versus HireVue. I remember going to many conferences where HireVue was in the room and they were talking about their software. Again, I am not picking on any software. I'm picking out for you the 6 top litigations that have occurred based on AI, solely based on AI, not based on the system that we already use in terms of applicant tracking or ATS system, not based on that, based on what the AI is doing. So this case was based on facial and video analysis. The AI was utilized to look at facial expressions to generate an employability score. So let me digress for one second. I just came back on last Wednesday from speaking at South by Southwest in Austin, Texas. When I was at South by Southwest, I was in a room with people developing AI software. I ran a core conversation where we got HR leaders, AI developers, AI designers, and HR tech, meaning technology and HR together in one room to have a conversation. It was probably the most meaningful thing that I've done in a long time around what does the future look like, because we all got in the room and, and talked about what the future looks like. But just as soon as I left that room, I went into another session and there was a person there who was pitching a piece of software that he had designed and got it funded for where he was going to use voice, a person's speech, to determine cultural fit in the organization. Let that sink in for a second. He is developing a piece of software based on speech. So what happens if I stutter, if I have speech impediment? And he's saying, he's pitching this to say speech can determine cultural fit in an organization? If I have an accent, if I speak with a brawl, if I sound like I come from the country or another country, how does that affect cultural fit? So AI developers are designing things they think HR wants, and there will be HR people who think that they need to go buy these things. Because they're trying to short circuit the process. All of this is going to result in us being in the middle of litigation. There's nothing else for us to be but inside of litigation. So with HireVue, it was a behavioral penalty. The black box behavior scored, it basically violated the ADA. You can't measure a human suitability through psychological or physiological data without triggering massive accessibility violation. If I have a stroke and my face doesn't look the way it's supposed to look, how do you determine— or if I have MS or some other thing that's wrong with me, or I had an accident and I have some disfiguration, We have to be way more conscious of what we are allowing inside of our HR processes. So there's Amazon, where Amazon was building an algorithm for performance and promotion. Now, we still, as HR practitioners, we still have AI software that is determined flight risk, that is built into your HRIS system, that is going to flag someone as a flight risk, that system may be flagging based on some pre-programmed notion that is in the back end. It may be flagging based on something that we don't even know. It could say, oh, a woman that's 37 years old who hasn't had a child or just recently got married, she's probably a flight risk because she's gonna go out and have a baby and never come back. We do not know because we're not asking the right questions what is actually in the backend of the AI softwares that we're putting into our organization. So Amazon built this software and then realized how discriminatory it was, and they had to shut it down because it discrim— it systematically excluded pregnant workers, disabled workers, and women who wanted favorable schedules and so on from promotion. So this use case was outsourcing bias and feeding objective productivity data into an ungoverned model that then created systemic discrimination. Okay, um, so they shut it down. Then this week, um, we had a couple cases. Two weeks ago, I posted a, um, a post on LinkedIn about chatbots and your managers using AI on their phone to do write-ups. There are two cases, US v. Hepperman and Warner v. Gilbargo. All right, the second one I'm going to talk about— there has been a rash of attorneys using AI to write the brief that they submit to court. And the AI that they use is hallucinating and making up cases and citations that are not valid. There have been several attorneys sanctioned and reported to the bar, and in this case, the attorneys were, um, sanctioned $15,000 each, and the case that they filed was dismissed automatically and they cannot refile the case. So when we think, oh my gosh, we have to have governance around our vendors, our providers, and how they use AI when they are representing us in our building and outside of our building. But the first case, US v. Heparin, Heparin, Hepner is the big one. Okay, so let me explain that one to you. He was being charged with fraud. The federal government was gonna arrest him. Before the government was going to arrest him, he went into Claude and he entered all the things about his case. He did that before he called an attorney. Didn't call an attorney. He entered all of his stuff and he used Claude to get legal advice. And Claude gave him a lot of legal advice. He then turned around and printed it all out and sent it to his attorney and said, this is what I want you to do. A judge ruled that the entire discussion he had with the AI was discoverable and needed to be turned over to the government. As discovery. Now let's bring it into the workplace. Many of you have auto chatbots. Auto chatbots, chatbot joints, okay. Auto chatbots is also discoverable. If an auto chatbot comes into a meeting where you are writing up an employee and the auto chatbot enters that meeting, everything you do inside of the meeting is discoverable and needs to be turned over to your attorney if there's litigation, which means you have a preservation of evidence. And I can tell you from my days in risk management, there's a thing called spoilation of evidence. This is risk management terminology where you have destroyed the evidence that is needed in litigation, and it's called a spoliation. I'll give you an example. I handled a case where a gentleman fell into a pothole for one of my city and municipal clients, and the client went ahead and filled the pothole right away. And the lawyer for the guy on the bike couldn't go and take pictures of the pothole, and so we were charged with spoliation of evidence, and we basically lost the case. Let me take that same analogy into AI. If your managers have their phone out and they're writing up Jane, Jane is a problem employee, I don't know how to talk to Jane, Jane is really not a good person, I'm not sure what to do, give me some language on talking to Jane. It's on their personal phone, they're using Gemini, Claude, um, ChatGPT, Grok, anything on their personal phone, that conversation is now admissible. Your attorneys are going to start asking, and you need to start asking your managers, when you wrote this person up, did you use anything else to write them up, or did you write them up out of your own brain cells? Because if the manager says to you, I used AI to write this person up, you have a problem. The, the case, Heppner, as I butchered the person's name, broke all the rules. Conversations with AI about terminations, internal investigations are not privileged. Most of you have Copilot on your desk. The only reason you might get away with it with Copilot is because Copilot is considered an enterprise application, and in the ruling, the judge says if it is an enterprise application, an argument could be made that it is protected because it's inside of the enterprise. I don't want to hold a bowl of water out in the sun based on that. I think we have to be way more conscious of what we are doing. I had a— when I posted this on LinkedIn 2 weeks ago, I had 365 HR people comment. All of them were in shock because they have been allowing chatbots into their write-up conversations with employees because it was automatic. It's there.

Ta-da!

So we can't say we didn't know how it worked because that's not a legal defense anymore. Because this case blew it wide open. So the last case is another big one. This one was filed exactly a month and a half ago. So litigation is continuing to evolve. The litigation is against Eightfold, which is another AI-powered hiring tool. Again, I'm not picking on any company. I am basically giving information based on going out on the internet, reading the cases, pulling it down, and then distilling it down to a PowerPoint slide for all of us. With EightFolds, their AI— and this one is a little bit, um, problematic. The hiring tool compiled and evaluated the applicant's data. It looked at work history, projected, um, career trajectory, culture fit, other personal traits. It assessed them. It may have pulled Consumer Reports and other things out in the ethers, public knowledge about the applicant. And it compiled it into a score, and then it accepted or rejected the applicant or gave them a low score, so you rejected them. This has now been certified as class action, another class action. So we have Mobley v. Workday, first class action. Second class action The Eightfold case is now being accused of violating the Fair Credit Reporting Act. And here's the thing, I don't think anyone that used the software knew it was exactly what it was doing. Hence my terminology at the very beginning about the black box. We do not know what our AI is doing because we don't ask the questions about what our AI is doing.

Doing.

Why all of this matters for HR: If courts agree that AI hiring scores are consumer reports, many HR teams will suddenly have FCRA-style obligations when they rely on external AI scoring services. HR will need clear vendor contracts, disclosures, authorization process and an adverse action workflow, whether to secure, um, some kind of indemnification or to prove that that was not a part of your final decision on a hire/no hire. So for all of us, and I can tell you because I I love AI. I'm not an anti-AI person, so I wanna set that record straight right now. I'm not an anti-AI person. I use AI all the time. I teach AI in instructional design for ATD. I teach AI in talent development for ATD. And so, and I teach AI across talent cycle. And I also teach AI governance. Across the talent cycle. I'm not anti-AI. I want you to understand that the tools that we have that we think are being used for expediency are now creating additional work for us. So I want to stop for a second and go into the chat and hear from all of you. So, There's a question, how to conceive of guardrails in the wild, wild west? Yes.

Yes.

How to get guardrails going? Absolutely. People using Copilot and other things without any usage policy in place?

Yes.

Not only usage policy, so let me, let me, set up a set here. I just said that I teach the program AI in Instructional Design. We get a lot of folks who are not in instructional design attending our program. It's a program I facilitate through ATD, Association for Training and Development, and I have been for— I've been a facilitator for them for many years and doing this program for the last 2 years now. What I find is that when people come to the program is they do not have a basis on how to use Copilot or any other AI that's been given to them. The AI popped onto their desk. They— it's there for productivity, but we have not given them the training for efficiency. And so we will spend 4 hours going through what is the proper way to prompt, and we will be training talent development instructional designers to go back to the organization to teach folks how to use Copilot effectively so that it is efficient. So we can't just go Oh, we've given them a tool. The learning curve for the tool is huge. All right. So keep putting your questions into the chat. I want to hear from you. I wanna customize this as much for all of you so that you walk away with some solid things to do. In every single case that I just went through, the organization relied on 3 defenses. The vendor certified it was fair. The vendor told us the algorithm was okay. IT approved the software. We didn't know how it made decisions. In every single case, the court ruled against that premise. The premise of we didn't know is not a premise. We have to ask questions, and for all of you attending, I will be sending out a list of questions that you should be asking your AI vendors. It was far too much to put it in here, but there are lists of questions like, what data was used to train the algorithm? How did you train the algorithm? What is the decisions that it is making? Walk me through the decisions step by step. Tell me when an applicant comes in, What's step 1? What does it do? And a lot of software vendors will say, well, that's proprietary. Well, guess what? Proprietary doesn't get a check from the vendor is not on the chopping block. And I want to just reiterate, just take this one step further. Many of us have this myth that IT governance is what governs us. The system is perfect because IT said it was. IT is only looking at the code. They're looking on at encryption. They're looking at security. They're looking at uptime. They're looking at does it validate within a software architecture. They are looking at the code. Does the code have any maliciousness inside of it? The system can be perfectly technically 100% putting out exactly what you need while in the background it's legally discriminating. And we have to understand that. There's two separate things going on at the same time and two things can be true. We can have a great system and it's also discriminating. So two things can be true at the same time, and we need to really understand that, um, that two things can be true at the same time. So for HR governance, there's a couple things that is required. Now, there's no specific guardrail right now for HR governance, which is why I wrote the book. There isn't one. There's not something that you can go out and pick up and say, here is the governance design for HR. There is lots of governance design for IT, but it is not the same for us. We have to be able to validate knowledge quality. One of the critical things about AI is that it is there to please you. It will replicate what you give to it. If our data that was given to the AI to make decisions from, we never hired a woman in any executive role, the AI will think that women are not qualified beyond a certain point. If you have, let's say, 10 bands within your payroll cycle, band 1 to band 10, and women stop at band 6, and you have an AI that's looking at performance, It will not assume that women belong in bands 7, 8, 9, and 10 if your prior data does not show that you have any women in those, those bands. Okay, so we have to validate the data quality. We have to validate what it's doing. We have to audit people decisions, and we also have to have bias mitigation. What are the people decisions that the AI is making? Then how does it validate its human outcome? There is a process that I developed inside of my book called epistemic governance, meaning that we look at the entire system that the AI is within and we walk the system as if we're walking a pathway. The AI step 1, this is what it does. Step 2, it goes here. If this happens, then it goes here. If this happens, it goes here. We can't demo software anymore based on the shiny red ball. I just wrote an article that is on our AI Governance for HR Substack, and the article is, if we put AI into HR and we decide it's ineffective, how do we undo it? That's a part of governance. When we put the implementation in, what happens if that company that sold us the AI goes out of business? How do we unwind the AI from what we have? What is the unwind for the AI, right? We have to ask those questions. Does the, does the algorithm systematic, systematically exclude protected classes? We have to ask the question, and it is up to us in HR to have an answer for that question. So for those of you in the EU, European Union, Australia, South America, Brazil, in Southeast Asia, All countries outside of the US have laws that govern AI. I was gonna create a slide to show you how many laws are in the US. Let me give you just an example. Right now, Texas has an AI law, New York, Chicago, Chicago itself, not the state of Illinois, Chicago, Colorado, California, Arizona is looking at some stuff. There's other states that are putting in— Florida tried to pass a law, but it didn't, it didn't make it through committee. So what we have in the US is a patchwork of AI governance laws that if we are doing business in those particular states, we have to know what those laws are. I will send a, a follow-up to this webinar with a list of all of the states and the state laws that are out there right now and how they impact us as HR leaders. So we don't have a global process. The European Union put in the EU AI Act, meaning the European Union AI Act. It says that any, um, AI used in HR is considered high risk. What is happening is individual states in the US is reading the EU AI Act and they are pulling pieces of it out and enacting it in their state or county governments. It will be a nightmare if we don't get an overall governance. The prior administration, not our current president, but the prior president, had a framework for an AI governance process. It did not make it into committee in Congress or the Senate. The current president last week released a framework saying this is a framework that we need to have for AI at work and AI in general. Will that make it anywhere? I doubt it because the appetite has to be in the Senate and the Congress to bring forward a mandate around governance for AI in the US. So for those of you in the European Union, Article 26 is what governs AI for us in HR. It says that the AI vendor cannot have a black box around how the AI functions when it is affecting talent decisions. Article 26 says, as the deployer, meaning you the employer, you are considered the deployer, you are responsible for what the vendor has sold you. So you cannot take the marketing brochure as the gospel. You must know what is there. I personally think that US employers need to learn the EU AI Act, which we are developing a course around that for HR, which we will have in, um, available in April. Um, I think we need to know the EU AI Act because the patchwork of things that are going on in the US is basically mimicking the the EU AI Act. It's, it's basically the same thing. They're just rewording it and removing the EU part of it. So deployer accountability says one thing: the legal buck stops with the entity using the tool to make the decision. You cannot outsource liability to your software vendor. The extraterritorial trap, meaning that a system that processes a single candidate located in the EU must be governed by the Act. The penalty for non-compliance is €20 million or 4% of your global annual revenue. For those of you doing global HR, you just got a new complication. This is no different to GDPR, the Privacy Act that came out a few years ago. I think US employers felt, oh, the Privacy Act is not going to go into effect. It did go into effect, and we've all complied. The EU AI Act will go into effect, and we will have to comply if we are going to operate in the global workforce. Um, any questions on that, please put it in the chat. All right, so one of the safety nets that we have counted on is employment practices liability insurance. And if you remember me saying at the very beginning, I come out of a risk management landscape, and so insurability and insuring risk is at the top of the things that I do. The absolute exclusion is now a new terminology that is being written into employment practices liability policies. Basically The insurance companies are saying AI has the ability to duplicate and replicate fast, and we are not covering that replication. So if you do something wrong, you're on your own. This is why governance becomes our requirement for long-term sustainability in the HR process. So I want to play this video for you, and I downloaded this this morning at 7 AM from Instagram. I take no credit for this video. The video was put up on Instagram by The Reluctant Accountant. That is her handle, and I want you to hear this, and I'm hoping this all works. So let's go.

That you don't have to— immigrants aren't coming to take your jobs. They could do that from their own country. So imagine your best employee never takes sick day, always online smashing deadlines, but they don't exist. And the salary you're paying them might be funding a foreign regime. What's going on? So as you can see from the headline, it says fake workers from North Korea use AI to exploit European companies. Let me tell you what's happening. There's a mini army of like North Korean operatives that are using AI to get hired by Western companies, including in Europe and the UK. We're talking fake identities, stolen LinkedIn profiles, AI-generated faces in interviews and even real people standing in for them on Zoom. Sorry, I know it's not funny, but anyway, in the US alone, over 300 companies were infiltrated. This generated at least $6.8 million for North Korea. So let me give you the quick, like, Netflix Heist movie version of this. So step 1, they had identity theft. They basically hijack real LinkedIn accounts, or they pay people to use their identity. And then step 2, they have this, like, AI enhancement So they use these AI tools to write the perfect CVs. They generate like culturally accurate names. They pass interviews using deepfakes with video filters. Basically, they look, sound, and write like a legit UK-based software engineer. Step 3, they have an interview hack where if AI isn't quite enough, they will hire a real person like a facilitator to do the interview. Step 4, they have this like laptop interception. The company obviously will send them a laptop, but they accept in. They log in remotely. Now they're inside your system. And step 5, they scale the operation. They have one person, multiple jobs, multiple salaries, and rinse and repeat. And some of these operatives are among the best performers in the company. So some people are saying this isn't just fraud, this is state-backed economic warfare. North Korea is basically bypassing sanctions, earning foreign currency, and potentially accessing sensitive company and the scary part is that recruitment has actually never been treated like a security risk. Companies built systems to stop hackers, not fake employees. This means that hiring could become a national security risk, and HR are going to be like the first line of cyber defense. It just goes to show that AI is not just a tool, it could be a weapon as well. Also, they're just messing it up for the rest of us. Like, we want remote work, but they're like, oh yeah, they're taking it too far. Like, there's always one employee that messes up for everyone else. This time it just happens that they might be in North Korea. Not gonna lie though, everyone else, this time it just happens that they might be in North Korea. I'm not gonna lie though, some of you might prefer to have some fake colleagues rather than the idiots in your office.

But I had to end on the joke. Are we prepared for this? I want to hear from you in the chat. Um, put in the chat what you think about this. Just wow. Gulp. Okay, so now do you understand why I feel beyond the discrimination because I am working inclusivity, my skills in risk management are now intensely necessary. And giving you that skill is going to be absolutely an imperative. And this is what we have— 300 companies in the US. They think there's about 400 or 500 companies. This is all legit. So what we thought we were doing in HR is not what we're really doing in HR anymore. So we have to shift. We cannot be passive consumers of the, of the tools anymore. The questions we ask— is the tool efficient and did legal approve the contract— is not the right question. The question that we need to ask right now is, does the system know what it claims to know, and can I verify this independently? We have to demand proof of fairness and actively Architect epistemic safety. Epistemic meaning we have to look at systems. We have to think in systems. In HR, we've been, we've been taught to think in process. As a risk manager, I think in systems. I used to work in factories where we would automate a line, and I had to figure out how long it would take an employee to put a widget into the next widget. And what was the safety concern of the employee putting the widget into the next widget. This is who we have to become. I want to give you some takeaways and tools. Governance is not negotiable. It's not non-negotiable. It's not something that we can put as a policy in the handbook. Governance is an actual day-to-day requirement. It is not a committee that you set up. A committee is essential, but it is not just the committee. We need named accountability. For every AI that is being used in your organization, a single person must be responsible for that AI tool. A single person must know that AI tool in and out. There must be data and, and data captured on what the tool is doing. We must look at AI outcomes. Did I get only women in this batch of applications that went through the tracking system? Who got excluded? What step is this AI taking that it shouldn't be taking? We also need to be independently verifying the vendor claims, meaning what is the demographic data that was trained on the AI? And let me explain why this is the most important step. Softwares get updated all the time. 15 minutes before this program, there was a Zoom update. I had to shut my computer down, accept the update, and I'm sitting here going, it's 4 minutes to time to start. It's not done updating, right? But it was lagging. So I had to do it. You have updates that will occur all the time inside of your softwares. Those updates reset the data. You need to know if it's updated, did it reset our data? Is it continuing where we left off? Is something new happening? So as the models iterate and get better, things happen. Many of you are deploying AI agents within the process. I was at an event last week where an individual was talking about she's going back to her organization to hire agents because they need a statement of work, and she's there to build out training the agents. We need evidence trails. We have to maintain documentation that will survive any hostile federal audit or legal audit. We need human in the loop. People must be able to override the AI at all points. This is not like traditional software where we can turn it off. The vendor that puts it in needs to show us how to kill it. And I know that word sounds like weird, but it is called a kill switch in AI. I've done my homework. Yes, it's called a kill switch, right? And we have to have continuous oversight, monthly cadence, metrics that we are following, looking for outliers. And we have to be able to put that in place now. One of the things that I did, and it's a part of the book that I'm releasing, my book will come out, um, right now we're shooting for May 20th. Uh, it might be a week or two later than that, but it should be out no later than June 1st. I developed what is called the SimplyFocus 5-step AI governance process for HR. I built out what is called the AI Failure Web, which is a diagnostic mapping that you use inside of your organization. And then I created a mitigation workflow. What happens if we find a problem? And then we have building governance architecture, which is how do we name ownership for the AI in our organization? And then the knowledge protocol, which is training your teams on what to look for, what to spot, how to assess it, how to move it from point A to point B. And then finally, an operational cadence. Here's what happens. Many of you will call your attorney. Your attorney will tell you, write a policy. A policy is not operational. The young lady from the video, she just described an entire hacking process. How would your hiring managers or your recruiting team be able to spot that if you don't have an operational cadence around being hired? How would they spot that? Today I gave you the why that you need all of this. To protect your organization, you must implement governance, but you must also have everyone on your team trained to understand the implications of governance on a whole. We have to walk our leaders through the litigation so they understand, don't pull your phone out and start typing stuff in. This is a part of where we are now. This is a part of who we must become. It is a lot of work and I hate to say this, AI in all aspects of the organization will create efficiency, We may even use it to terminate people. I'm not going to be not honest about that. But the governance requirement is not going to go away. And if we deploy AI into our organization and we don't have an exit strategy, then we have a problem. I've created an Intro to AI Governance program It is a microlearning. There's 60 minutes of video content walking you through the first, um, I'm walking you through the first two phases of SimplyFocus, the Failure Web Analysis and the Mitigation Workflow. Those two parts of the process, the governance architecture and so on, is much more advanced. Everyone has to start with a web analysis and then a mitigation workflow. When stuff happens, what do we do? When the software gets updated, what do we do? How do we know what it's doing? If we can't analyze and then create mitigation, there's no point of us creating a committee to have a governance committee, right? There's, there's some steps before the governance committee. So we have the Intro to AI Governance program. I've also created a collab where monthly— it's a, it's a monthly program where we'll all get together and just talk about what is going on inside of AI in HR. And then we also have the AI Governance for HR Substack where I post articles literally every single solid day. Into it. You can go to our website courses.inclusionlearninglab.com and all of this information is there. I will put a link into the chat before we end. So I, I'm offering everyone a $20 off coupon for the live program and you're able to register for the live program for $129 instead of $149. This is a fundamental program and a great place for everyone to start learning about AI governance. Any questions about what I've covered so far? Any questions as we wrap up today? Yeah, it's scary to know that this is happening. I feel the need to educate everyone and that is part of what what we're doing here. And we will do these programs at least every week or every other week to educate everyone on what this is. But the hard work of building governance is not quick and simple. It is work, and we have to do that work every single day. So you can connect with me on LinkedIn. Um, I would love it if you would join our Substack, AI Governance for HR. And, um, one second, I've just added, um, I've just added our— I think it's courses. Um, yeah. Yes, so, um, any questions from anyone that attended today? Any questions that you have about the content? Did I leave anything out? What else do you need? Um, please let me know. Send me an email, um, and let me know. So thank you again for attending this program. We are so happy you're here, and stay tuned for additional governance workshops that we put on at least once or twice a month. Take care, everyone, and have a wonderful afternoon. Thank you. Take care.