Georgia AI Work Product: New Rules for 2024

AI is everywhere in our legal work now, and it’s creating real headaches for work product privilege. These tools are getting smarter, helping us with everything from legal analysis to accident reconstruction, but we have to figure out how to shield those outputs from discovery. For those of us handling complex cases, like the big wrecks we see in Roswell, Georgia, this isn’t some law review debate. Whether the work an AI does for you is protected directly affects how you build your case and what you have to turn over.

Defining Work Product in the Age of AI

Georgia’s work product privilege, which you’ll find in O.C.G.A. Section 9-11-26(b)(3), exists to stop the other side from piggybacking on your hard work by protecting materials you prepare for litigation. It’s always covered things like my notes and internal memos. But AI throws a wrench in it. Say I use an AI to analyze crash data from a wreck on Holcomb Bridge Road in Roswell, is the AI’s vehicle dynamics report protected like a memo I typed myself?

Judges everywhere are struggling with this question. It boils down to whether the AI is just a fancy word processor or if its work actually shows my thinking and legal strategy. A 2026 report from the American Bar Association showed just how split the courts are. Some see AI as part of the team, doing what the lawyer tells it to do. Others worry about the “black box” problem, that the AI is just crunching data on its own, not really capturing an attorney’s work. I’ve been doing this a long time, and from where I sit, if I’m the one crafting the prompts, feeding the AI specific information, and then analyzing its output to build my case, that’s my work product. End of story.

Working through the “Ordinary Course of Business” Exception

The “ordinary course of business” exception is a big trap when it comes to protecting work product AI. Any report or analysis that a company does as part of its normal operations isn’t privileged, even if a lawsuit pops up later. Think of a standard incident report written right after a slip-and-fall at a Roswell business, that’s probably discoverable because they do it every time, litigation or not. So when does using an AI stop being routine business and start being litigation prep?

Let’s say a Roswell trucking company uses an AI to monitor its fleet data all the time, just for general safety. That’s business as usual. But then one of its trucks gets into a huge wreck on Mansell Road. If their lawyer then tells the AI to run a new, specific analysis of that one crash to figure out liability, that’s different. The purpose changed. If you, the attorney, are using the AI specifically to map out your legal strategy or find holes in the other side’s case, your argument for privilege is strong. But you have to document that direction. If you don’t have a paper trail showing you ordered that analysis for litigation purposes, the other side will just call it a routine business record and get it.

Roswell Accident Cases: Specific AI Applications and Privilege

Roswell accident cases can get messy, whether it’s a pileup on GA-400 or a slip-and-fall in the Historic District. AI can be a lifesaver, chewing through mountains of data from traffic cams, witness interviews, and vehicle black boxes. For instance, an AI could create a physics model of a t-bone crash at the intersection of Alpharetta Street and Woodstock Road, giving you a detailed breakdown of the impact forces and potential injury mechanisms.

Using AI for this kind of deep analysis means you have to be thinking about protecting legal privilege from day one. It’s not enough to just push a button on the software. You need a system.

• Attorney Direction: I have to be the one telling the AI what to look for, what questions to answer, and what data to use. The record needs to show my brain is guiding the process.
• Human Review and Interpretation: The AI’s output is just raw material. My job is to take that output, analyze it, mark it up, and decide how it fits into our case strategy. My notes and conclusions about the AI’s report are pure work product.
• Confidentiality Protocols: We have to treat these AI reports like any other confidential litigation file. If one gets accidentally emailed to opposing counsel, the privilege is likely gone.

Even the State Bar of Georgia is telling firms to get their act together, putting out guidance that you need a formal, written policy for using AI in litigation. That policy needs to cover everything from how you pick the AI tool and feed it data to how you review the results and keep them confidential. If you don’t have these policies in place, good luck arguing privilege when the other side sends a discovery request for your AI analysis.

The Substantial Need and Undue Hardship Standard

Just because something is work product doesn’t mean it’s locked away forever. Under O.C.G.A. Section 9-11-26(b)(3), the other side can still get it if they can show a “substantial need” for it and prove they’d face “undue hardship” trying to get the information any other way. So how does this play out with AI?

Picture this: I use an AI to build a simulation of a disastrous multi-car pileup on Highway 92 by the Chattahoochee River. If the AI used public data and their expert could (eventually) build a similar model, they probably can’t show undue hardship just because it would cost them more time and money. But what if my AI used proprietary data they can’t get? Or if the simulation is so complex that no human expert could possibly replicate it in time for trial? Now their argument for needing my AI work is much stronger. This is also where they’ll scream about a “black box” if the AI’s process is a secret, arguing they can’t challenge my expert if they can’t see how the AI came to its conclusions, which might persuade a judge to order disclosure. Being transparent about the AI’s general process (when you can) can sometimes be the best defense for the privilege itself.

The issues go beyond just work product and bleed into attorney-client privilege and expert reports. What happens if I give my client access to an AI tool to help them gather information for their case? As long as that AI is clearly a channel for providing them with legal advice, like an advanced paralegal, those communications should be privileged. But if the client starts using that same AI tool to do things unrelated to their legal case, that protection is probably gone.

The rules get even more specific for expert reports. Georgia’s discovery rules, specifically O.C.G.A. Section 9-11-26(b)(4), dictate what you can discover about the other side’s experts. If your testifying expert relies on an AI to form their opinion, you can bet opposing counsel will demand to see the data you fed it and how the AI worked. This makes the line between a testifying expert (whose work is largely discoverable) and a non-testifying consulting expert (whose work gets much stronger protection) incredibly important when AI is in the mix. Firms have to be very deliberate about using AI for internal case strategy versus using it to prop up a testifying expert’s opinion.