AI Workers Comp Claims: SC Puts Accuracy on Record
South Carolina's June 17 bench order puts AI claim work on the record. For contractors, the risk is not the tool. It is the claim fact that gets repeated.
AI workers comp claims became an audit issue in South Carolina when the Workers' Compensation Commission issued a June 17, 2026 bench order requiring accuracy and caution with generative AI work product (SCWCC, June 2026). For contractors, the premium risk sits in claim summaries, reserves, compensability notes, and return-to-work narratives that feed the EMR.
AI workers comp claims stopped being a software story in South Carolina on June 17, 2026. The South Carolina Workers' Compensation Commission published a bench order advisory and a four-page Full Commission order that says the quiet part clearly: accuracy still belongs to the person filing the work (SCWCC, June 17, 2026).
The useful line for contractors is short: "shall ensure the accuracy of all work product" (SCWCC, June 17, 2026). It belongs in your renewal meeting.
AI can help compress a messy claim file. It can also flatten the fact that changes your Experience Modification Rate (EMR, also called the mod). A summary that makes a return-to-work offer sound clean, leaves a reserve looking settled, or treats a disputed body part as accepted can travel farther than anyone intended.
The National Council on Compensation Insurance (NCCI) says experience rating analyzes an employer's payroll and loss data, usually the latest available three years, against similarly grouped employers to calculate the mod (NCCI ABCs of Experience Rating, 2025). That mod adjusts manual premium. Claim file facts are not side notes. They are premium inputs.
Why AI workers comp claims create EMR risk
The South Carolina order doesn't say AI is useless. It says generated content can carry inaccuracies, bias, cybersecurity vulnerabilities, intellectual-property issues, and privacy concerns (SCWCC, June 17, 2026). That is exactly the problem with claim summaries.
In our reviews of Southeast contractor worksheets, the expensive files often have a boring origin. A lost-time claim gets summarized once, then that summary follows the file into reserve reviews, broker notes, counsel updates, renewal submissions, and the NCCI worksheet.
South Carolina's order also reaches lawyers and parties before the Commission, not just Commission employees (SCWCC, June 17, 2026). It tells lawyers they may not delegate professional judgment to an AI system under Rule 1.1 and Comment 8 of the South Carolina Rules of Professional Conduct (SCWCC, June 17, 2026). That matters because legal analysis and claim posture can shape whether a file is viewed as compensable, disputed, closing, or still worsening.
The $75,000 file can echo for three years
NCCI's experience period is generally based on three years of payroll and loss data, but it can include up to 45 months depending on policy timing (NCCI ABCs of Experience Rating, 2025). NCCI also notes that the current policy is not used because the insurer is not required to report that data until 18 months after policy inception (NCCI ABCs, 2025). That lag is where a stale narrative can become premium.
Use NCCI's own roofer example. A $200,000 roofing payroll at a $63.171 rate per $100 of payroll produces $126,342 of premium before clerical payroll, and the combined example shows $126,867 of total premium (NCCI ABCs, 2025). NCCI then applies a 1.25 mod to produce $158,584 of modified premium (NCCI ABCs, 2025).
Now put the article's $75,000 lost-time claim into that kind of account. If an AI-assisted summary helps keep the claim in the wrong posture and the mod moves by five points, $126,867 of standard premium changes by about $6,343 in one year. Across three affected rating years, the swing is about $19,030 before taxes, assessments, schedule credits, and carrier pricing. The claim doesn't need to be wrong by $75,000 to cost real money.
Where the claim story bends
The risk is not a robot making a final decision. That is too neat. The risk is a confident paragraph that makes an open medical issue sound closed, a light-duty offer sound refused, or a reserve look like an unavoidable outcome rather than a snapshot.
The SCWCC order is direct on filings: a lawyer may not submit AI-generated authority, factual assertion, evidence, or argument without first reviewing it for accuracy and legal support, and the order cites Rules 3.1, 3.3, and 8.4(d) for false statements and fabricated citations (SCWCC, June 17, 2026). Contractors should read that as a claims lesson. If the fact is important enough to file, it is important enough to verify before it reaches the mod.
Return-to-work narratives are the easiest place to lose the plot. Construction files move fast. Superintendents change jobs, restrictions get revised, and a short-term no-duty period can harden into a long-term disability story. AI is good at smoothing that mess into a paragraph. Smooth is not the same as accurate.
What an audit would check
An audit checks whether the claim story that moved through the carrier, third-party administrator, broker, and counsel actually matches the values and status used on the experience rating worksheet. It tests whether the reserve, injury type, recovery posture, and return-to-work narrative still support the mod being charged. It also checks whether any South Carolina claim touched by AI-generated work product has a human-verified record behind the facts that matter.
If AI touched a South Carolina claim file that is still inside your experience period, send us your NCCI worksheet and we'll review the premium impact before the next renewal.