Insurance carriers are now using artificial intelligence to process workers’ comp claims, and it’s a huge problem for injured workers, especially anyone with a pre-existing condition. These algorithms are full of AI bias, automatically flagging and denying claims that a human would approve. As these automated systems make more and more of the decisions, we’ve had to get a lot smarter about building our workers’ comp evidence to fight back. The question is, how do you challenge a denial that was spit out by a computer and win a fair settlement for your client?
Key Takeaways
- Get ahead of AI denials by documenting the claimant’s entire medical history, especially pre-existing conditions, using detailed reports from specialists and clear diagnostic images.
- Bring in a vocational expert. They can put a real number on how the new injury impacts the client’s ability to earn a living, providing a human story and hard data that an algorithm simply can’t ignore.
- Don’t be afraid to go after the AI itself. Subpoena the insurer’s system parameters and the data it was trained on to expose any built-in biases against specific medical histories or groups of people.
- Attack the AI-generated report in court. Argue that it shouldn’t be admitted as evidence because it’s a black box, it’s likely biased, and it completely fails to consider the unique details of your client’s medical situation.
- Dig in for a long fight. You need to build a rock-solid evidence file that proves medical causation, even with a pre-existing condition, and that almost always means getting an expert to testify specifically on the aggravation.
| Factor | Warehouse Worker Case | Data Entry Clerk Case |
|---|---|---|
| Claimant Age | 42 years old | 55 years old |
| Injury Type | Lower back injury (lumbar aggravation) | Bilateral carpal tunnel syndrome |
| Pre-existing Condition | Degenerative disc disease | Mild, intermittent wrist pain |
| AI Denial Reason | Not major contributing cause | Idiopathic, exacerbated by age |
| Settlement Amount | $125,000 | Not specified |
| Resolution Timeline | 14 months | Not specified |
Case Study 1: The Warehouse Worker and Lumbar Aggravation
We just had a case with a 42-year-old warehouse worker, Mr. David Chen, that shows exactly how this plays out. He hurt his lower back lifting cartons at a Fulton County distribution center near Hartsfield-Jackson Airport in late 2025. The problem? He had a known history of degenerative disc disease, a classic pre-existing condition. The insurer’s AI flagged it instantly and denied the claim. The denial letter basically said the AI decided his work injury wasn’t the main cause and that this was just his old condition getting worse on its own.
But the injury itself, a sudden, bad twist while lifting, was completely different from anything in his old medical files. Our whole battle was proving the work incident aggravated his degenerative disc disease and made it disabling. We knew what we were up against: an AI that just looks at historical data and can’t tell the difference between a condition slowly getting worse and a specific, traumatic event that blows it up. They design these systems for speed, not for the kind of careful thought a real doctor would apply to a case like Mr. Chen’s.
So, our strategy was to build a medical record the AI couldn’t argue with. We sent Mr. Chen for an independent medical examination (IME) with a top orthopedic surgeon who knows spines. This doctor put the pre-injury MRI from 2023 side-by-side with the post-injury one from early 2026, writing an expert report that pointed out the exact changes to the disc herniation and nerve impingement that were clearly new or made much worse by the lift. We followed that up with a functional capacity evaluation (FCE), which gave us objective numbers on his reduced lifting ability and range of motion, tying his new limitations directly to the 2025 injury, not the old condition.
Then we went on the offensive. We sent a discovery request demanding to see the details of the insurer’s AI, specifically its training data and how its algorithm handles pre-existing conditions. Of course they resisted, claiming it was proprietary. We fired back, arguing the AI’s decision-making affected Mr. Chen’s due process rights, after all, O.C.G.A. Section 34-9-100 guarantees every injured worker a fair process. That pressure, piled on top of our medical evidence, was enough to make them blink. With a hearing looming at the State Board of Workers’ Compensation, they came to the table and we settled for a $125,000 lump sum. It covered all his medical bills, lost pay, and permanent partial disability. Getting that done in 14 months after an AI denial is pretty quick work.
Case Study 2: Carpal Tunnel Syndrome and Repetitive Motion
In another case, Ms. Sarah Jenkins, a 55-year-old data entry clerk in Midtown Atlanta, came to us after developing bad bilateral carpal tunnel syndrome in late 2025. She’d had some mild wrist pain before, nothing major. But the carrier’s AI system denied her claim, calling her condition idiopathic (meaning, “we don’t know the cause”) and blaming it on her age instead of her job. This is a classic AI trap. The software has no idea how to handle injuries that build up over time from multiple factors.
We’re right by the Fulton County Courthouse, and we knew right away this was a textbook example of AI bias. The algorithm probably saw her vague pre-existing symptoms and the absence of a single “accident” and just assigned a low score for work-relatedness. So we hit them from two sides. First, we got an affidavit from Ms. Jenkins’s supervisor that spelled out just how much typing and mouse work she did all day, every day, critical workers’ comp evidence. Second, we got the objective medical proof: an electrodiagnostic study (NCS/EMG) that confirmed she had severe nerve compression.
We then brought in an occupational medicine specialist to connect the dots. He testified that her specific, repetitive job duties were the direct cause and aggravating factor, pushing her from “prone to it” into full-blown carpal tunnel syndrome. He explained that her job was the specific event that tipped her over the edge into a painful, disabling condition. That’s the kind of sophisticated medical reasoning that these AI programs just can’t process (at least not yet).
Staring down a hearing with an Administrative Law Judge at the State Board of Workers’ Compensation, the insurer finally agreed to mediation. We hammered them on how their AI was completely unqualified to judge a complex repetitive injury case. Their rep kept talking about how “efficient” their system was, but he couldn’t deny the strength of our medical and work evidence. We settled the case for $80,000, which paid for her surgeries, all the physical therapy, and the TTD benefits she needed while recovering. All told, it took about 11 months from the claim being filed to get that check, with a lot of that time spent fighting after the initial denial.
Case Study 3: Shoulder Injury with Rotator Cuff Degeneration
Here’s another one. Mr. Robert Miller was a 60-year-old maintenance tech at a Gwinnett County plant. In early 2026, some machinery shifted on him, he strained his arm overhead, and wound up with a full-thickness rotator cuff tear. The carrier’s AI took one look at his file, saw some degenerative changes on an old image from a 2024 physical, and immediately denied the claim. The algorithm’s logic was simple and wrong: he’s older, he had some degeneration, so the tear must be from age, not the accident. This is what these pattern-matching systems do, they mix up correlation and causation all the time, especially with older workers.
We had to prove the clear line from that specific accident to the acute tear, despite his pre-existing conditions. We started by getting the employer’s incident report, which backed up Mr. Miller’s story about a sudden, forceful strain. Then we sent him to an independent orthopedic surgeon who issued a critical opinion. He stated that even though Mr. Miller had some underlying wear and tear, the way the accident happened was a textbook cause of a traumatic rotator cuff tear. The doctor made the key point that people walk around with asymptomatic degeneration all the time until a specific trauma, like Mr. Miller’s, turns it into a real, painful injury. That’s a distinction an AI is going to miss every single time.
Then, just like in the Chen case, we subpoenaed the documentation for their AI. We wanted to see exactly how it weighed pre-existing conditions against a one-time trauma. Once we forced them to produce the documents, we found the smoking gun: the algorithm was heavily programmed to blame pre-existing conditions and was terrible at analyzing a single traumatic event unless it was something catastrophic. This became an incredible piece of workers’ comp evidence that proved their system was biased from the start.
We took that finding and immediately filed for a hearing, specifically calling out the insurer’s use of a biased AI. The last thing they wanted was to have their precious AI’s methodology picked apart in a public legal hearing, so they quickly got serious about settling. The case closed for a $185,000 lump sum, covering Mr. Miller’s surgery, PT, lost wages, and permanent impairment. It took 16 months, but it showed that you can beat the machine if you attack it with solid, human-focused medical and legal work.
Handling workers’ comp claims now means we have to be constantly on guard. You have to build every case assuming an algorithm is going to scrutinize it, and you have to be ready to expose the biases in that system. You need to know the medicine, but now you also have to understand the tech, and its weaknesses, to get your clients what they deserve.
What do you mean by “AI bias” in a workers’ comp claim?
It’s when the software insurance companies use to review claims has built-in prejudices that cause it to make bad decisions. The AI might be programmed, intentionally or not, to automatically deny claims for people who have pre-existing conditions or are in a certain age group. It looks at historical data patterns instead of the facts of your specific injury, leading to unfair denials.
Why does having a pre-existing condition make an AI claim harder?
Because an AI sees a pre-existing condition as an easy excuse to deny the claim. The algorithm is trained on mountains of old claim data, and it learns that claims involving pre-existing conditions are often denied. So it automatically assumes your new injury is just the old condition acting up, not a new work-related injury. It’s on us to bring overwhelming medical proof to show the work incident is what actually caused the current problem.
What’s the best evidence to fight an AI denial?
You need a combination of human expertise and hard data. This means getting reports from independent medical specialists (IMEs), comparing new MRIs or X-rays to old ones, and having an expert doctor testify about how the work injury aggravated the old condition. We also use functional capacity evaluations to get objective proof of what you can and can’t do now. And sometimes, the best evidence is information we subpoena about the AI system itself to prove it’s biased.
Can you really subpoena the insurer’s AI system?
Yes, absolutely. In Georgia, we can use the discovery process to demand that the insurance company hand over information about its AI, including its programming, the data it was trained on, and how it makes decisions. They’ll fight it and claim it’s a trade secret, but judges are starting to understand that if an AI is making legal decisions about a person’s life, we have a right to see how it works.
What does Georgia law say about pre-existing conditions?
Georgia law (specifically O.C.G.A. Section 34-9-1(4)) is clear: if a work accident makes your pre-existing condition worse, it’s a compensable injury. The key is proving the work incident was the “proximate cause” that aggravated the old condition. The whole fight, especially when an AI is involved, boils down to proving the work injury is what triggered your current disability, not just that you had a bad back or wrist to begin with.
