---
id: "workshop:crafting-ai-use-policies"
type: "InnovateUS workshop"
title: "Crafting AI Use Policies"
url: "https://innovate-us.org/crafting-ai-use-policies"
date: "2026-05-20"
duration: "60 minutes"
series: "AI for Public Sector Legal Professionals"
instructors: ["Marlene Biener","Melissa Szozda Smith"]
moderators: ["Mina Noroozkhani"]
recording: "https://www.youtube.com/watch?v=7Fn9voErstU"
transcript: "automatic, from the recording"
source: "InnovateUS (innovate-us.org)"
licence: "CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/), as stated on innovate-us.org"
---

# Crafting AI Use Policies

InnovateUS workshop · 20 May 2026 · 60 minutes

This session provides guidance on developing robust policies for AI usage within a government setting. The session provides a framework for structuring a comprehensive AI use policy, including security, privacy, transparency, equity, accountability and oversight, and effectiveness. Through case studies and policy examples, attendees will learn how to draft comprehensive AI use policies that protect clients, increase efficiency, and maintain public trust.

## Learning goals

1. Draft a comprehensive AI use policy that addresses security, privacy, and equity considerations specific to a government setting.

 2. Apply real-world case studies and policy examples to identify best practices and common pitfalls when implementing AI governance frameworks.

 3. Balance operational efficiency with public accountability by building oversight and transparency mechanisms into AI use policies.

## Speakers

- **Marlene Biener**, Chief Operating Officer, Association of Prosecuting Attorneys
- **Melissa Szozda Smith**, Assistant Section Chief, Consumer Protection Section, Office of Ohio Attorney General

## Moderators

- **Mina Noroozkhani**, Deputy Attorney General, Office of the General Counsel, California Department of Justice

Part of the series [AI for Public Sector Legal Professionals](https://innovate-us.org/workshop-series/ai-for-public-legal-professionals-series-).

Recording: https://www.youtube.com/watch?v=7Fn9voErstU

## Transcript

Automatic transcript of the recording: names and terms can be misheard, and the times can be a few seconds off. Each time links to that moment of the video. The transcript alone: https://innovate-us.org/crafting-ai-use-policies.txt (plain text) and https://innovate-us.org/crafting-ai-use-policies.json (every segment with its start and end second).

**Mina Noroozkhani**

[0:00](https://youtu.be/7Fn9voErstU?t=0) Thanks, Agueda. It's great to be here with everyone today and also to moderate this important topic, which is how do you craft AI use policies that make sense for those of us who work in the public sector? So we're hoping that our discussion today is gonna be valuable to government employees who may be in the process of drafting their generative AI use policies or are about to do so. We're gonna start with some brief introductions so everyone who is attending understands our own experiences with generative AI in our offices. I'm a Deputy Attorney General in the California Attorney General's Office. I help to develop our generative AI policy and also the workflow that relates to the review and approval of Gen AI tools within our office. I'm gonna turn it over first to Melissa and if she's joined us, Marlene, to give a brief overview of their background with Gen AI in Ohio and New York. Melissa, why don't you start?

**Melissa Szozda Smith**

[0:50](https://youtu.be/7Fn9voErstU?t=50) Thanks, Meena. Appreciate it and I appreciate being here today. So my name is Melissa Smith. I am the Assistant Chief in the Consumer Protection Section with the Ohio Attorney General's Office. My role with the office in this space has been a little bit just on demand. So essentially what had happened is about three years ago, Ohio was one of the first Attorney General's Offices to have a policy related to AI and specifically related to Gen AI. As we'll discuss moving forward, it's a very broad policy and the main thing is it does allow for certain uses of it, but it places responsibility with the user at the end of the day. So that is a policy that we have now, but now we're also looking forward and that's kind of where I come in to say, what are the considerations? What are the factors? What are the tools? And who do we really need to have in the room when we're making these decisions on crafting a policy moving forward?

[1:54](https://youtu.be/7Fn9voErstU?t=114) So that's kind of a longer way of saying Ohio actually has a policy, but as we know everything with AI is ever changing. And so we're also looking forward down the road at how we are going to adopt new policies and implement those changes.

**Mina Noroozkhani**

[2:13](https://youtu.be/7Fn9voErstU?t=133) Thanks, Melissa. I don't know if Marlene has joined. I don't see Marlene. So I think we're gonna go. Okay, so we're gonna go ahead and then just get started. And when Marlene joins, maybe she can give a brief introduction. So we've collected a series of questions to help you learn from our collective experiences in drafting these policies and also address some of the challenges that we faced during the drafting and implementation process. And as Agueda said, there's gonna be an opportunity for Q&A at the end. So please just put your questions in the Q&A box and we'll get through as many of them as we can. So let's start with a general question for you, Melissa, which is how has government legal work distinct from private sector legal work when it comes to AI risk? What in your view is at stake that makes this conversation different in a public agency context?

**Melissa Szozda Smith**

[3:05](https://youtu.be/7Fn9voErstU?t=185) I think there is so much to be said here and it really spans across the spectrum. So I think, first of all, I think there's still some fear and uncertainty surrounding AI. And so when we talk about a government organization implementing AI, there's obviously fear on that, but I think to the general public, if they know government's using AI, there's almost this point in time where it's a real inflection point to say, oh boy, if the government's doing it, it might be here to stay because as all of us on the call likely know, due to budget reasons and other reasons, the government sometimes isn't quite the first adopter of things like this. I think layered on top of that is the sense of responsibility that we have to the public. We are there to serve them. And so there's public perception involved in that as well as a heightened sense of responsibility in my opinion.

[4:05](https://youtu.be/7Fn9voErstU?t=245) Going along with that is we are the stewards of a lot of data. Our office has so many different data points for people, some that we have and some that's voluntarily turned over to us. So there's another sense of responsibility. And then I think that what's the final horseman, I guess I would say, and that is the actual laws and statutes and things that the government is subject to that a private business might not be. So when we talk about AI, you can't have this conversation without talking about public records, without talking about privacy, given the personal information that we have. So there are all of these things layered on top of one another that I think if a private business is doing something, having an initiative, a policy surrounding AI, it's a little bit, well, actually a lot different than when the government does it.

**Mina Noroozkhani**

[4:59](https://youtu.be/7Fn9voErstU?t=299) So let's talk a little bit about the actual drafting process. So if you were advising a legal office starting from scratch, what are the two or three provisions you'd say are absolutely non-negotiable in any AI use policy and why? And I'll actually start answering from my own experience working on this policy for California. I would say there are a couple of things you really need to make sure you address. The first is the actual vetting and approval process. So in other words, who is gonna review the tools for approval? Is it legal? Is it security, human resources, unions? And how is that review process going to work? And the second is deciding what actually constitutes responsible use of generative AI in your office? And how do you plan to train employees on that responsible use? What do you think, Melissa?

**Melissa Szozda Smith**

[5:52](https://youtu.be/7Fn9voErstU?t=352) Yeah. I agree with both of those points. Often times when we discuss our policies or be in it, you know, or what is no, if someone's sitting in a deposition, one of our employees, how can we defend our policy? So I think you've really hit on some of the most important parts is who moving forward is responsible for implementing the policy, for being consulted when we're talking about a policy, and then what does the ultimate responsibility look like at the end of the day from the employee? The other big picture thing, it wouldn't necessarily go into a policy, but I think it's so critical to understand, and okay, we're going to create a policy, but what problem are we trying to solve?

[6:54](https://youtu.be/7Fn9voErstU?t=414) And I think that it's like that for all of AI, whether it be one specific use case or one specific tool, it's certainly very easy to get squirreled away because there's so many cool things having to do with AI, but how does our office want to use this? How are we trying to solve it? What are we writing a policy to solve? And I think I have some good news, which is that it looks like Marlene has joined us. So Marlene, if you're there, if you want to come off and introduce yourself.

**Marlene Biener**

[7:23](https://youtu.be/7Fn9voErstU?t=443) Hi, yes, and I'm one, so very sorry for the technical difficulties I had joining. I think that may be as a testament to technology and all of the woes that we're facing with this, but thank you all so much for having me. I'm so glad to be a part of this conversation. I can tell you a little bit about myself. I am Marlene Biener. I am the Chief Operating Officer at the Association of Prosecuting Attorneys. APA is a national nonprofit organization. We work with prosecutors all over the country. We provide training, technical assistance. We work on policy development, and we work with offices of all sizes. And it's really on a whole host of issues, but one of the main things I'm focusing on is of course the use of data and AI and all of the emerging technology that prosecutors are working with and are seeing on a daily basis. And so in my role, I work directly with prosecutors, other justice system partners, and this is such a big part of the national conversation right now.

[8:28](https://youtu.be/7Fn9voErstU?t=508) And I am just so grateful to be here before joining APA. I was a Deputy Attorney General at the New Jersey Attorney General's Office, so also come having worn that hat as well. So I appreciate being here today. Thank you.

**Mina Noroozkhani**

[8:42](https://youtu.be/7Fn9voErstU?t=522) Thanks, Marlene. So we were actually talking right before you joined about if you were advising a legal office starting from scratch, what are the two or three provisions you'd say would be non-negotiable in an AI use policy and why? So what are your thoughts on that question?

**Marlene Biener**

[8:58](https://youtu.be/7Fn9voErstU?t=538) Yeah, and so I think that there's several. So I think one of the major things as a government agency, as a prosecutor's office, as an attorney general's office, one of the things that you really need to do is make sure that you have very clear rules about what can be entered into these AI tools. I think that is just a baseline starting from scratch. A clear office policy should really have staff understanding what is prohibited. Is the information confidential? Is it sensitive? Is it regulated? Is it case-specific information? And I think the second kind of part to that is these approved AI systems and making sure that, so the one piece is the personally identifiable information and what information gets put in, but also second to that is also what systems are approved and making sure that you understand as a staff member by reading this policy, what is appropriate, what is inappropriate. And I think a good rule of thumb is if this is not something that you would post on a public, in any other public space, this is maybe not something that you should be putting into a public AI tool.

[10:03](https://youtu.be/7Fn9voErstU?t=603) So I think that's one really important bucket. I think a second is on having a clear vetting and approval process. So I think the policy should make very clear that employees who are using AI tools, they might be already, they might be using a tool because it looks helpful. They might be using a tool because they don't necessarily think that it is something that has any sort of public facing component to it. But I think at a minimum, there needs to be some version of an approval process that is inputted into any AI policy that really has those different considerations in mind. I think somebody who has that sort of professional responsibility lens would really need to be a critical component of that vetting process. And the last thing I'll add is I think the human verification, the human accountability component of AI use is a really essential part of any policy. I think a policy needs to state very clearly that this is a tool.

[11:03](https://youtu.be/7Fn9voErstU?t=663) It does not replace professional judgment and it's responsible use remains the responsibility of the user who is using it. I think any AI tool really needs to be reviewed for accuracy, reviewed for bias, reviewed for completeness or even relevance, which is something that an AI tool might not always get right. And I think especially for attorneys, really just understanding that your professional license is the thing that you need to really be keeping in mind when you use these tools. And I think a policy really needs to create that impression that it is a tool, even if it's an approved tool, it is still a tool and it does not replace that human verification and accountability.

**Mina Noroozkhani**

[11:48](https://youtu.be/7Fn9voErstU?t=708) Thanks, Marlene. Those are great tips. So let's talk a little bit about security and privacy. And I know Melissa, you touched on this a little bit in the government context, but what are your views on the top security or privacy considerations when drafting a generative AI policy? So Melissa, why don't you start?

**Melissa Szozda Smith**

[12:04](https://youtu.be/7Fn9voErstU?t=724) So Marlene just touched on it a bit ago, which was, so for example, in Ohio, we have Copilot and that's our secure system. And so we have, through our policy, the ability to put confidential information in there because it is a closed program. But it needs to be made very clear that there is a difference between our enterprise's Copilot GCC product and ChatGPT or Gemini or Quad, or whatever public facing tool there is. And so I'm not sure how many people are familiar with Copilot, but what they always say is if you have a little green check, you can put in there confidential information. Sometimes what people don't equate is, okay, I'm going to go home and I'm on my personal machine, but I'm going to use some version of Copilot that's actually not our entity level one. So that can be a little bit tricky. So from security, it's really working with IT to understand if you are going to secure these products, number one, is it actually secure for us to be putting this confidential or proprietary information in?

[13:13](https://youtu.be/7Fn9voErstU?t=793) And number two, equally as important, is how can we educate our staff on when these tools are appropriate to use? On the flip side is when are tools inappropriate to use? What information absolutely cannot go into a ChatGPT because actually in Ohio, we don't completely prohibit ChatGPT. It's my understanding some offices do. We just say only certain information can go in there. So if you want to ask it a legal research question, go for it. If you need to put very specific details, that's going to need to go into Copilot. And then from a privacy perspective, obviously there are a ton of privacy laws out there that are, some are at the federal level, some are at the state level. So it's really important for each office to know, you know, number one, the information and data that their offices have on a, likely on a section by section basis, division by division, and then to appropriately know how to address the issue of privacy.

[14:24](https://youtu.be/7Fn9voErstU?t=864) So, you know, in our office, it was interesting. The beginning I started saying, I'm in the consumer protection section. Something I didn't say is my main legal focus is on consumer data privacy. So that's sort of how I got looped into this. That doesn't make me a chief privacy officer for our office. And so there again, we see, we need to know who to pull in at every step of the way, but because our offices are entrusted with so much information, it's really important to hit those privacy considerations and concerns on the front end when you're contracting, as well as through education.

**Marlene Biener**

[15:00](https://youtu.be/7Fn9voErstU?t=900) I think it's important to understand what AI environment we're talking about, I think understanding what you're using it for. If you are using it for that very sensitive government information or that case level data, I think that's a very different consideration. And I think that's where you're really going to get into the different privacy and security considerations of public consumer AI tools versus these enterprise or these closed system AI models. And I think that's what a lot of government agencies are really focused on, is those closed system, those enterprise systems. And I think knowing the difference between all of these different, because it's so overwhelming, there are so many different systems, there's paid versions, there's free versions. So understanding what is the security risk that you might be putting out there if you are using these tools that are not fully closed or that don't necessarily have that enterprise level security. And I think you also hit on it, but I think not all AI use carries that same level of risk. So if you want to create, I don't know, like a training kind of thing for your office or model your questions or like a moot, I think that that is something that AI can, you know, you can kind of use things that are a little bit more, have that lower level of security, but it's really, it's because you're not inputting that information that's extremely sensitive. And so understanding what you're using the AI for and kind of having those considerations. I think one additional thing I will add is that some of the products that are already existing in offices across the country are now, the newer versions of them are already using embedded AI. And so

[17:02](https://youtu.be/7Fn9voErstU?t=1022) your office might already have some of these embedded AI tools and things like Adobe or Microsoft Word or Zoom, even that don't necessarily have the policy in place by your office. So just kind of thinking about what potential tools you already have in place that weren't procured after sort of the development of this AI policy, but are already existing within your office. And what is the sort of comfort level with the security measures within those different tools that are already existing within your office that are not necessarily the public consumer level AI, but something that your office has essentially inherited this tool that now has these new features?

**Mina Noroozkhani**

[17:43](https://youtu.be/7Fn9voErstU?t=1063) Thanks, Marlene. Yeah, we're actually going to talk about that a little later when we talk about the procurement problem. So as we know, writing the policy is only one piece of the puzzle. Getting it actually adopted is another challenge entirely. So who needs to be at the table during the drafting process to ensure adoption is a success? And Melissa, you touched on this briefly in your response to the last question. Do you want to expand on what you mean?

**Melissa Szozda Smith**

[18:10](https://youtu.be/7Fn9voErstU?t=1090) Yeah. So, you know, it's interesting because I think who needs to really adopt it and implement it and put the stamp of approval are going to be your higher-ups in administration, and you need to get buy-in from them. The flip side of that, though, is because AI is so multifaceted, it can be very easy for even those higher-up members to really want to have a say in it. And as I'm sure anyone who's interested in AI has done, you can go down a million rabbit holes. So I do think it's important to strike a balance between who are going to be what I'll call the boots on the ground and then who you need buy-in from. And interestingly, you know, in every decision we're making regarding AI, we find that a new person needs to be brought in, whether it's someone from administration or someone from, you know, they're ones you're going to know. You're going to know legal. You're going to know IT. We're going to bring in procurement. We found there are varying levels of criminal tools versus civil tools. What about communications when you have to try to educate staff on how this is going to go? So there are a lot of people that you need at the table. I think ultimately, for me in an attorney general's office, the highest person that's going to have buy-in who needs to be convinced is the attorney general himself, especially because this could be a very large budgetary responsibility for an office in general.

**Mina Noroozkhani**

[19:48](https://youtu.be/7Fn9voErstU?t=1188) Okay, great. Marlene, what are your thoughts?

**Marlene Biener**

[19:51](https://youtu.be/7Fn9voErstU?t=1191) Yeah, I completely agree. I think that office leadership is so critical, not only to be at the table, not only to provide input on risk and risk tolerance and, you know, the different tools and priorities of the office, but I think also to really let the staff know that this is a priority, show that this is something that is important and it is enforceable and it is something that needs to be complied with. And I think office leadership can really push into why this policy is so important, what responsible use looks like, and that compliance with it is expected. I think that another piece of this is HR because, you know, at the end of the day, this is an employment, you know, workplace policy. And so what does it mean if a person does not comply with the policy? What does that look like from an employment perspective? How is this enforceable from that HR perspective? And I think two other things I'll add here. One is training staff. If your office has a sort of training component to it, the training staff should be in the room, I think, to ensure that there is that training, an educational component of not only the inner workings of the policy, not only the governance structure and how to be able to ensure that these tools are being used appropriately, but actual implementation.

[21:20](https://youtu.be/7Fn9voErstU?t=1280) Because somebody who might not be familiar with any of these tools, you now have information about these approved tools. I think there is a really important component to it of educating and training the staff on how to use these tools safely and effectively and appropriately. I think they're innovative. They're exciting. They could potentially create efficiencies. But I think it's just so helpful to have those folks kind of provide that information about how to use them well. And I think the last thing I'll add is, if you can, adding folks who are going to be the ultimate end users of this policy and these tools would be helpful. Different types of staff who might use the policy and the tools in different ways, it might be really helpful. If you have a data analyst, maybe they're going to use this tool, or an investigator, they're going to use it a little bit differently than a lawyer in your office would. And so, maybe having a representative from those different perspectives or user groups would be really helpful to ensure that the policy makes sense when it's operationalized.

**Mina Noroozkhani**

[22:28](https://youtu.be/7Fn9voErstU?t=1348) So, related to the question that we're answering now, and Melissa, you touched on this when you said that the AG needs to sign off on these policies when you work for an AG's office. How do you guys recommend managing the complex landscape of legal, IT, HR, elected officials in getting a policy approved? And Melissa, why don't you start?

**Melissa Szozda Smith**

[22:49](https://youtu.be/7Fn9voErstU?t=1369) Yeah. So, I think something that is very important, and, you know, I think at the beginning we forgot, but we're always supposed to give this disclaimer. These are my views, not that of our office, but this truly, truly is my view 100%. I think it is critical to have a person in charge who can be held accountable, who can touch all the different sections or divisions, and really steer the ship on this. So, you know, a question that I might ask IT, maybe, can we do something? And IT can say, yes, this meets our security requirements. But then a question I'm going to turn around and ask legal is, well, can we legally do this, and should we do it? And then a question that I might have for our team that interacts with the public more so is, can we do this? How is this going to be perceived from the outside? So, having one person in charge to touch on all of those points and then funnel it up the chain is really important. And I think that as much as we talk about this as implementing AI, to a certain extent, it is also implementing change management, and it is a huge project for the office. So, in order to get all of the parties within the office on the same page, I personally would advocate for one or two point people in charge who know all the different divisions to go to, and then ultimately can take that to the AG or to whomever is, you know,

**Mina Noroozkhani**

[24:22](https://youtu.be/7Fn9voErstU?t=1462) in the AGs about AI. I think that's a really important point about having a single person or at least two people who know everything about it and can funnel those decisions up.

**Marlene Biener**

[24:34](https://youtu.be/7Fn9voErstU?t=1474) Marlene, anything you want to add? Yeah, I just, I love what Melissa shared. I think that everybody who's doing this or who's sort of responsible for their own lane is going to have such a rich and varied perspective on AI. And, you know, they're either going to be caring a lot more about procurement or a lot more about the ethical, you know, and legal considerations. So, it is so important to have that singular person or group that can really focus in on putting all of that together into a cohesive policy. I think it's also really important to ground a policy in a shared set of, like, principles or, you know, before talking about specific tools and their functionalities, I think really understanding, you know, sort of on baseline, like, what are we trying to do here? What are we trying to accomplish? What is the mission and vision for this particular policy? What is acceptable? What is not? And sort of kind of thinking about that and also not forgetting sort of like the culture of the office and the culture change component of it is so important.

[25:40](https://youtu.be/7Fn9voErstU?t=1540) And I think the final thing I'll add is this is such a balancing act because the technology is just evolving so rapidly. The use of the technology is just, it just is across the board. And so, when you have something that's evolving so rapidly, you know, wanting to be careful, wanting to move thoughtfully, but not wanting it to be so restrictive that it almost, people won't follow it because they're just already using these tools in their everyday life. And so, just sort of that, having all of those different considerations in mind, I think is a really helpful starting point.

**Mina Noroozkhani**

[26:20](https://youtu.be/7Fn9voErstU?t=1580) So, as we know, policies, and you just touched on this, Marlene, often lag behind technology. So, the question for us is, how do you draft a policy for tools and uses that you may not have actually anticipated yet and ensure that you're not having to rewrite the policy every few months? And one thing I'll say, speaking from experience in California, was we tried to ensure the policy was broad enough to address how we plan to vet tools and how to responsibly use them. But we also require that we develop a policy specific to each Gen AI tool that we approve. And so, that way, we can ensure we're addressing the specific requirements that may apply when using one tool, but not another. For example, what data can safely be input into a tool versus what data can safely be input into a tool versus another tool? Melissa, what are your thoughts on this question?

**Melissa Szozda Smith**

[27:10](https://youtu.be/7Fn9voErstU?t=1630) I think that's exactly how it has to be done. Like I said, in the Ohio Attorney General's office right now, we have a very broad policy, and we're currently trying to figure out how these might apply or how to apply the policy to tools in the future. The state of Ohio, so under the governor, has its own policy. And that policy, what I thought was unique, is it lays out the guardrails around using AI, but then it says for each specific use case, we need to submit it to, I think they call it the AI Governance Council. And that council can then look at specific tools and apply them, or I'm sorry, approve them on a case-by-case basis. And I think a policy like that gives a little bit more wiggle room where we're kind of saying, we're encouraging you to come up with new use cases, but we really need to know what they are before you launch them. We need to approve them. And so, you're approving almost in real time when it comes up versus a blanket approval or a blanket disapproval.

**Mina Noroozkhani**

[28:18](https://youtu.be/7Fn9voErstU?t=1698) Those are great points. How about you, Marlene?

**Marlene Biener**

[28:20](https://youtu.be/7Fn9voErstU?t=1700) Yeah, no, I think you both stated that really well. I think I kind of keep coming back to de-emphasizing particular tools or particular products and really focusing in on principles or risk levels and workflows. And I think that it is such an important way to sort of think about this because, like we mentioned, the technology just developing so rapidly. I think that there's also just new court opinions or so many things that are also other elements, external elements that might factor into a policy that is not just the rapidly evolving technology or the use of it. So, I think one thing that has to also be really, and I know we'll talk a little bit more about governance structures later, but I think that that is just such a critical part of it is understanding that this is an ongoing process, that it's not really going to be a one and done policy, but really thinking about having those regular review periods and being able to have that cross-functional oversight with that individual or group that is going to be operating the whole thing and really having that mechanism in place for updating guidance as all of these things continue to evolve and emerge.

**Melissa Szozda Smith**

[29:42](https://youtu.be/7Fn9voErstU?t=1782) And if I can just add one thing, because I think that was a great, great summation, is in crafting this policy specific, I think you just have to remember the saying, don't let perfect be the enemy of good. And so, no one is ever going to craft the perfect A policy out of the gate because the technology is changing so quickly. So I think it's so critical to focus on what's the best that we can do for right now. Like Marlene said, with the understanding we probably are going to need to revisit this, but let's get the basics in place and then we can take a deeper dive in the future.

**Mina Noroozkhani**

[30:23](https://youtu.be/7Fn9voErstU?t=1823) Yeah, I completely agree. And once you actually draft the policy, we know there's this whole process for actually implementing it. So let's talk about the steps that we all recommend when taking a written policy and actually putting it into action. And I'll start the discussion. So in my view, I think there are two key steps. The first is you really need to create some sort of generative AI committee or working group that is responsible for reviewing and vetting the tools that people want to use. And as we've all discussed on this call today, you need to ensure the right people are sitting at the table. So it might be legal, information security, procurement, human resources, and probably someone from executive leadership. And then the second key thing is actually creating a workflow for that committee. So how are employees in your office going to submit requests to use tools? Who is going to collect those requests? How are decisions going to be made about which tools to approve and which ones to deny? So Melissa, how has this worked in

**Melissa Szozda Smith**

[31:25](https://youtu.be/7Fn9voErstU?t=1885) your experience? It's worked pretty similar to what you just described. We have had an AI committee in place for quite some time that varies across IT, legal, library, litigation support, all these different people. And we are just now getting to this really critical point of trying to create the workflow of how these requests will happen because we were very focused for quite some time on getting a generative AI product within our office that could benefit the most number of staff members. And now what we're seeing is an interest, for sure, an interest in certain tools. And so we're at this point where we're kind of saying, okay, hold on to your horses for a minute, we get it. But we need to figure out how we're going to evaluate these tools from all the different angles that we've discussed a number of times. But I think having the committee is really important. The other thing that we've kind of what I'll call soft launched is we have, in our office, everyone has access to Copilot Chat. And then there are about 120 people who have access to a full Copilot license. And of those 120 people, we have around 20 to 30 who we call Copilot champions. And these are people who really are interested in AI. And they come to us and they say, you know, we wish AI would do this for us, or have you explored this tool or this feature. So we're really leaning in on our users now to understand how what the demand might be in the future, whether it be in a Copilot generative AI related space, or whether it be a completely different tool, because I was actually shocked at the level of interest, as well as the level of experience that people in our office have with AI, even some of them

[33:29](https://youtu.be/7Fn9voErstU?t=2009) just like as a hobby on the side, and they really serve to teach us a lot about AI, and to open our eyes to the direction that, you know, people are looking to use this in the future.

**Mina Noroozkhani**

[33:41](https://youtu.be/7Fn9voErstU?t=2021) I think that's a great tip leaning on power users in your office, whoever they may be. Marlene, what do you think about any steps you would recommend when taking a written policy and

**Marlene Biener**

[33:50](https://youtu.be/7Fn9voErstU?t=2030) putting it into action? Yeah, I think one of the critical elements of designing a good AI policy is thinking about how to operationalize it and roll it out. It's funny that you mentioned that, Melissa, we have we do do these hackathons with Microsoft, and we usually have prosecutors from across the country, some some from ag offices as well come out, and we kind of pull them in the beginning to think about like, okay, what is your sort of comfort level with AI, and what is your, you know, how do you feel about it, and there's usually a bit of trepidation, or a little bit of, you know, I'm not sure, and we, it's like a day training of understanding, prompting, 101, like thinking about, you know, developing it with real, you know, use cases, and inputting the information, and starting to build out some of these things yourself, and what we see is by the end of the day, there is such an excitement, there is such an appetite for using these things, and a lot more of a comfort level, so I don't even think it takes that much time to be able to help, you know, develop this out a little bit more in your office, and I think that the, you know, I said it before, but I think a training element of this, and how to use it is such a critical component of implementation, and I think that it's something that is, I think, going to pay dividends, because you're going to have this policy, now you have these tools that you've procured for your office, and now you have this excitement, and this comfort level in using it, and the ability to be able to use it appropriately, and then just one other thing I'll add, is I completely agree on the importance of these committees, and this sort of development of this robust internal governance structure, and I think this delicate dance of doing, creating a policy that is broad enough, so that it doesn't have to be updated every, you know, 18 days, or something, you know, just not something that has to be updated so frequently, but something that, you know, they can really be able to take, and understand what it is, and what they're supposed to do with it. I think there is potentially that broad interpretation, or that latitude, that you might be able to take with it,

[35:52](https://youtu.be/7Fn9voErstU?t=2152) as a person who sees that, and so I think the importance of having those, if you have questions, if there is something that's open-ended, that you're unclear about, just to be able to have that in place, so you know who to go to, in order to make sure that it's operationalized in the way

**Mina Noroozkhani**

[36:08](https://youtu.be/7Fn9voErstU?t=2168) that the policy was actually intended. So let's talk a little bit about cost, because obviously that's top of mind for people in the public sector, so how can we balance the demand for generative AI tools in our offices, against the time it might take to review them, and also the actual cost of buying these tools, and one thing I've personally found very helpful, is developing a matrix that helps you compare the utility of the tool with the risk of using it, so for example, is it a tool that's helpful to a handful of people, or to most of the organization, is it a low-risk tool that doesn't expose data, or is it a high-risk tool that has access to data across your organization, and then making decisions based on that matrix, about whether the tool should be front-loaded for review, or deprioritized, and Melissa, I know you touched upon this discussion of having power users who might recommend certain functionality, so maybe you can start the discussion and talk to us about balancing those interests against cost and other considerations in the public sector.

**Melissa Szozda Smith**

[37:13](https://youtu.be/7Fn9voErstU?t=2233) Yeah, so I mean, the reality is every office is working with a budget, and that can be very difficult, whereas we look at our friends in the private sector, and they might have a little bit more. The big thing I would say is when you receive a request for some sort of product, it's really important to see if a product that you already have offers that specific functionality, so for instance, although we've not turned it on yet, one of the most highly requested tools is call recording and call summarization, so in Ohio, at least the AG's office, we've not turned that on yet, but we can get that through copilot, so the question is why would we go out and buy a standalone tool for that, so we've seen through our team that we actually could get in things that we already have. The other thing is if you have a product like Lexis or Westlaw, that might be separate from what you're thinking are your big gen AI tools, but what are they offering? I was actually blown away this week, or last week, we had a training by Everlaw, and I've always used Everlaw for document review, but they were showing us some of their AI features that they now offer, so I would say that's a big thing, is look at the tool, look at the functions that the tools you already have offer, regardless of whether you have those tools flipped on or not, and then the other thing is you can spend a lot of time reviewing a ton of different tools. At some point, I think you just need to make a decision and go with it, because certainly it changes so quickly, and you'll see one big gen AI tool saying, oh, actually, our competitors is better as of this moment, and then next week, it's a different competitor, so you could play

[39:14](https://youtu.be/7Fn9voErstU?t=2354) that game all day long and test out products all day long, and then when you're at the end of 30 days where you've somehow managed to test five products, oh, guess what? Now the one you thought was the worst has this feature that now makes them the best, so I think going in with a common understanding that we need to focus on what's going to work best for our office right now, understanding, again, that it might not be the best 30 days from now, but then 60 days from now, it might be, so I think that's something it's tricky, because you think there's some finality, but there's really not.

**Mina Noroozkhani**

[39:56](https://youtu.be/7Fn9voErstU?t=2396) Marlene, what are your thoughts on this cost-benefit analysis that you have to do

**Marlene Biener**

[40:00](https://youtu.be/7Fn9voErstU?t=2400) in the public sector? Yeah, I think that what we're seeing across offices is just, you know, there is such a resource constraint, there is such an understaffing, so there are so many reasons why AI, I think, can seem like such an exciting opportunity for these offices to potentially create efficiencies. I think that the tools are sometimes being, the interest in these tools is maybe also being created a little bit faster than the opportunity to review them, like you both mentioned. I think that in terms of efficiencies, it is an interesting thought, and I think you both sort of hit on, like, what is the need for this tool and really having that sort of understanding of what is this tool going to do that we don't currently, you know, have the ability to do, and what problem is it going to help us solve, you know, things like that.

[41:02](https://youtu.be/7Fn9voErstU?t=2462) But I think that that all has to be balanced with, is this actually going to take a lot of staff time to implement? And also thinking about, it's essentially, sometimes it could be considered a substitute, right? Like, instead of creating a summary from scratch, for example, maybe this tool will draft the summary for you, but now the trade-off is you have to edit it, you have to review it, you have to potentially, you know, do a lot more of editing of it. And I think that that is something worth considering, because something that might feel like a huge efficiency might almost be like a substitution. And maybe it's not, maybe it is an incredible tool and it is super effective, but I think that's something to consider, is it's not necessarily that clean substitute. So thinking about all of those things, the staff time required, training or oversight necessary to help really employ it. And I think you both really hit on the two major things, which is risk and value. And I think those are the two sort of pillars that you'll need to make sure that you're moving quickly enough so that you're not wasting so much time reviewing these, you know, larger number of tools on the market, but also doing your appropriate due diligence.

**Mina Noroozkhani**

[42:13](https://youtu.be/7Fn9voErstU?t=2533) So let's talk for a minute about what I'm going to call the procurement problem, which is, you know, you might have a great policy, but you still need to figure out how you're going to procure the tools and how you're going to vet the software that's coming up for renewal that might have Gen AI components. And I know Marlene, you referenced this earlier. So let's talk about how you can navigate that landscape. In California, one thing I've found helpful is developing a workflow that identifies all the ways a tool can come up for approval in your office. So for example, is it a standalone generative AI tool or is it existing software you already use that now has a generative AI component? And then based on that, how do you want to review use of that tool in your organization? So any tips that you guys want to share on deciding how to actually vet tools and reviewing them based on, you know, how they come up for review in your office? Melissa,

**Melissa Szozda Smith**

[43:08](https://youtu.be/7Fn9voErstU?t=2588) do you want to start? So this is something that we are actively working on in our office, trying to figure out how to best do this. And there are kind of two different components. The first is normally run through our ITS, but it's the ITS department of, you know, if someone comes to us and they say, I want a tool that can do call recording, whose responsibility is that to go locate that tool? Is it ITS? Is it the person making the request? And then obviously there are going to be all of the security privacy components layered on top of that. And then once we do locate that tool or identify that tool from our actual contracting side and procurement side is how are we going to draft the appropriate AI clauses into our contracts? And relatedly for our software that's already existing, whether we're using it for AI purposes or not, what is the vetting process to see what AI they have integrated? And we might not know about it. And if those tools should be turned off, I'm sorry, turned on, need to stay off, all of those things. So that is something that we are currently working through in Ohio. But what I think is great about conversations like this is it brings to the forefront the notion and the idea that these are things that we all need to be thinking about. I love to hear that California seems pretty far ahead in the game on that.

[44:39](https://youtu.be/7Fn9voErstU?t=2679) But it's going to be a work in progress, I think, for every office.

**Mina Noroozkhani**

[44:43](https://youtu.be/7Fn9voErstU?t=2683) Marlene, anything you want to add?

**Marlene Biener**

[44:46](https://youtu.be/7Fn9voErstU?t=2686) I completely agree. I think this is one of the most significant operational challenges of AI use in the workplace. I'll just add, I think, you guys said it so clearly that I won't have much more to add, but I do think that it's going to be a work in progress. Making sure you're also aware of the AI use, and I said this earlier, within the softwares and the products that you already have in your office, and understanding if those things now exist in the tools that you're using every day, and making sure, you know, asking those direct questions for those traditional software renewals as well to make sure that, you know, there is that risk consideration there too.

**Mina Noroozkhani**

[45:26](https://youtu.be/7Fn9voErstU?t=2726) So we have about 10 minutes left, so I'm gonna move forward with a few additional questions, and then hopefully we can get through some of the Q&A that folks have been adding. So let's talk for a moment about governments. So once you implement a generative AI policy, there's still this need to monitor use of those tools at your agencies, and ensure that data access remains restricted, people are using the tools responsibly, there aren't hallucinations in the tool, I could go on. So have you implemented a governance framework within your offices, or Marlene advised offices to do so, and what does that look like? Melissa, do you wanna start?

**Melissa Szozda Smith**

[46:07](https://youtu.be/7Fn9voErstU?t=2767) Sure, so we have not gotten to that point yet. I think the obvious first step is, you know something's not working if a court brings it to your attention. But, you know, as part of that too, I think that something that offices need to keep in mind is what are the courts expecting from AI now? What are, in Ohio we have different local rules at every which way you can turn. So certainly having an understanding of those local rules, and then making sure that your attorneys are complying with those. Now what's tricky is, unfortunately for attorneys, if you find out that the rule wasn't being followed, it's probably not a good thing. So it's kind of like, no news is good news, I suppose, if it's not being brought to our attention by a court. But certainly that is something that is very important, and I think will grow and mature as use of AI and further definition of the policy comes into being.

**Mina Noroozkhani**

[47:19](https://youtu.be/7Fn9voErstU?t=2839) Marlene, what about you? What are your thoughts on governance key steps?

**Marlene Biener**

[47:23](https://youtu.be/7Fn9voErstU?t=2843) Yeah, I think that it's one of the most critical pieces of this entire conversation, is really about the importance of a good governance structure. I think a written policy on its own, you know, offices can spend a really long time drafting a really strong policy, but I think it's so important to have a good governance structure behind it, because if it's inconsistently applied, or it's outdated in some particular way, or it's not aligned with the way people are actually using technology, it can kind of not be as effective as it really should be. And I think it creates the accountability within the office, it creates clear pathways to understand who is responsible for applying these tools, who is, you know, doing all of the things that we've kind of talked about today. How are those risks assessed? If there is a question, or if there is a practice that's not being, you know, appropriate for the way that the policy was written, like, what are the elements to that centralized oversight within your office?

[48:28](https://youtu.be/7Fn9voErstU?t=2908) And I think that because of the nature of AI, and the way it's being adopted, this is just such a critical component that can't be understated when developing an AI policy. I think that the governance structure within your office has to be sort of hands-in-hand with the policy itself.

**Mina Noroozkhani**

[48:46](https://youtu.be/7Fn9voErstU?t=2926) And Melissa, you touched on this a little bit, but, you know, we've all seen that there's this push to ensure that the use of generative AI is disclosed, and some courts are mandating disclosure when such tools are used. So how do you recommend handling these disclosure requirements?

**Melissa Szozda Smith**

[49:05](https://youtu.be/7Fn9voErstU?t=2945) I think that's really tricky because there are mandated rules by courts in Ohio, I'm sure throughout the nation, and I'm sure they all look very different, but, you know, the lawyer in me, when I see this disclosure always says, but disclose what? So to me, it is very different to create a brief using gen AI, not read it, and turn it in, and not disclose it, or even disclose it, versus I write something, or even use gen AI, but I go through it meticulously, I check for hallucinations, and then I run it through gen AI to check my grammar. But the problem is, and interestingly, I was just listening to a different CLE yesterday, in which judges are saying, but we don't want the latter of the two, but we've written the rule to cover all of it. Internally, this discussion came up, and someone said, I think there should be a disclaimer every time someone uses AI, and I said, I think that's crazy because we don't know in which capacity, and certainly, if I ask someone a legal question, and they, in 30 seconds, give me a legal answer, and say, but this is just what co-pilot said, I appreciate that disclaimer.

[50:22](https://youtu.be/7Fn9voErstU?t=3022) Do I care if they run their memo through an AI product to, again, grammar proof it, spell check, et cetera? No, not really. The other thing that I would say is when we talk about internal disclaimers, I would just be very leery of, I do think that our younger generation is more comfortable with AI, and not necessarily in a bad way. I think some of our more experienced attorneys and heads in our office are less experienced, and they have a tendency to trust AI less. And my fear is that if we're requiring disclosures on everything, that that somewhat gives an air of illegitimacy or the laziness to our younger attorneys, and it could really hamper them moving forward. So that's kind of my personal opinion on internal disclaimers, if we're making someone disclaim everything. But certainly, whatever the court rule is, holds, and it dictates, in Ohio, I would love to get a library put together internally of all of our court rules in Ohio, and I'm assuming that other states either are thinking about that or have already done that.

**Mina Noroozkhani**

[51:36](https://youtu.be/7Fn9voErstU?t=3096) Let's skip to our last sort of wrap-up question, because there is at least one question in the Q&A that I want to get to. So if someone in this workshop goes back to their office tomorrow, and they want to start the process of drafting or improving their AI use policy, what is the single most important first step you'd recommend? Maybe you can give us sort of your elevator pitch. Marlene, why don't you start?

**Marlene Biener**

[52:01](https://youtu.be/7Fn9voErstU?t=3121) Yeah, so I think it's important to level set and to understand how AI is already being used in your office, even if you don't have a policy, if you're thinking about starting a policy. AI is already there, it's already being used, most likely, by people within your office, most likely by many of the people within your office, even if they don't necessarily know that they are using AI. To Melissa's earlier point, I think that there is so much already in existence, and so getting a sense of who is using AI, how are they using it, I think that's a really helpful starting point before you actually start putting pen to paper, because I think there is so much experimentation with AI already, and so before starting from this sort of scratch, really getting a sense of what is your office culture, what is the sort of appetite, what is the excitement and interest level of your office, and how are they already using it? And I think it's okay to have that initial assessment, and for people to kind of speak openly about AI, and how they feel about it, and what they're doing, what they're using it for, and I think from there, the policy becomes much more specific to your office, it becomes much more manageable, because you're coming from this place of informed understanding of the way your office is already interacting with these tools, and then that can help kind of develop out the remaining elements of your policy.

**Mina Noroozkhani**

[53:29](https://youtu.be/7Fn9voErstU?t=3209) So I think in the last two minutes that we have, I do wanna get to at least one of these questions that one of our participants has put in the chat, and the question is, even within a closed AI system, how do you navigate conflicts and walls within public offices? So for example, one part of the office can't share information with another. So Melissa or Marlene, do either of you have thoughts on this issue?

**Melissa Szozda Smith**

[53:58](https://youtu.be/7Fn9voErstU?t=3238) I think it's a good thought. I think it's a technical thought that I don't have an answer to. Now, I'm sure whomever asked this will know enough to say, no, Melissa, you're wrong, but I will say it's been my experience in using Copilot that it is only retrieving from my, for me personally, it's retrieving from my inbox, from my Teams messages, from my OneDrive. I have not had the experience in which it's pulling from someone else's inbox, Teams, OneDrive, et cetera, and our case management system right now doesn't interact with Copilot. So that's not pulling from there. So everything else I think that it would pull from would sort of be public, I guess I'll say, but that's been my experience with Copilot.

**Mina Noroozkhani**

[55:01](https://youtu.be/7Fn9voErstU?t=3301) So I know there are other Q&A questions. We didn't get a chance to go through them, but thank you for everyone who put time and thought both into attending and putting those in there, and also to my fellow panelists, Melissa and Marlene. Agueda, I'll turn it back over to you to wrap things up.

**Agueda Quiroga**

[55:18](https://youtu.be/7Fn9voErstU?t=3318) Okay, thank you so much for the session. Amazing, and congratulations. Thank you so much for our participants. It has been really, really good one. We hope to have you again, maybe next term or something, to go deeper on this topic, which I think, as you have seen, there are so many questions. So thank you so much, everybody. Next Wednesday, final session in the series. We hope to meet you there. I'm going to close the session now. Bye, take care.

* * *

Original page: https://innovate-us.org/crafting-ai-use-policies · Licence: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/), as stated on innovate-us.org
