AI in Claims: The Rules Adjusters Now Work Under (and What to Put in Your File Notes)
AI is already in your claim files. It labels your photos, summarizes your notes, suggests line items and ranks which roofs need a person on them at all. Regulators have noticed. On June 12, 2026, the Texas Department of Insurance issued Bulletin B-0003-26 telling every regulated entity, and the third parties working for them, that AI-assisted decisions must meet the same legal standards as any other decision. TDI said it expects a person to review and agree with consequential decisions before action is taken.
Texas isn't alone. As of October 8, 2026, 29 jurisdictions have adopted the NAIC's AI Model Bulletin, and at least 13 states have guidance limiting how insurers use aerial imagery. Here is what the rules say as of early November 2026, what they mean for an adjuster's work, and what to put in your file notes.
General information, not legal advice. Policy language, state law and your carrier's guidelines control.
Texas: Bulletin B-0003-26
The TDI bulletin doesn't create new penalties. It reminds insurers that existing law already covers AI, and spells out what TDI expects:
- Same standards. Decisions made or supported by AI must not be inaccurate, arbitrary, capricious or unfairly discriminatory.
- Human review. For consequential decisions, a person should review and agree with the decision before anyone acts on it.
- Governance. Insurers should have oversight, risk management, internal audit and testing to catch errors and bias.
- Third parties are covered. Vendors and others working with regulated entities fall under the same expectations.
- TDI will check. It may ask about AI use during exams, in filings and when investigating complaints.
The bulletin cites several Insurance Code chapters, including Chapter 542 (claims processing and settlement) and Chapter 4101 (insurance adjusters). It also notes that Chapter 4201 bars using AI to make adverse utilization-review determinations. That provision applies to health, not property, but it shows how far some lawmakers are willing to go.
For an independent adjuster, the key point is simple. If an AI tool contributes to a scope, a coverage position or a denial, the licensed adjuster is still the one making the call, and the file should show that.
The NAIC AI Model Bulletin: 29 jurisdictions
The NAIC's Model Bulletin on the Use of Artificial Intelligence Systems by Insurers sets similar expectations: written governance programs, risk controls and oversight of AI systems and vendors. According to the NAIC adoption map, as of October 8, 2026 it has been adopted in 29 jurisdictions (27 states plus DC and Puerto Rico).
| Status | Jurisdictions |
|---|---|
| Adopted the NAIC bulletin | AK, AR, CT, DE, DC, HI, IL, IA, KY, MD, MA, MI, MS, NE, NV, NH, NJ, NC, ND, OK, PA, PR, RI, SC, VT, VA, WA, WV, WI |
| Own insurance-specific AI rules or guidance | California, Colorado, New York, Texas |
| Notably not on either list | Florida, Georgia, Louisiana, Minnesota, among others |
"Not on the list" doesn't mean "no rules." Unfair claims practices laws apply everywhere, whatever tool a carrier uses. It does mean the expectations are less clearly written down in those states.
More NAIC work is coming
The NAIC has been piloting an AI evaluation tool for regulators to use in market conduct and financial exams. Twelve states participated in the pilot, and adoption was targeted for the NAIC Fall National Meeting, scheduled for November 14–17, 2026, in Dallas. If it's adopted, examiners will have a standard set of questions for carriers about high-risk AI systems, and claims tools are likely to be part of that.
Aerial imagery: not the sole basis
Aerial and satellite imagery is the most visible AI-adjacent tool in property claims, and it has its own set of rules. A March 2026 analysis by Troutman Pepper Locke, first published in Carrier Management (republished by the firm), lists 13 states with bulletins or statutes on insurer use of aerial imagery: Alabama, Delaware, Louisiana (by statute), Maine, Maryland, Massachusetts, Michigan, New Hampshire, North Carolina, Pennsylvania, Rhode Island, Tennessee and West Virginia.
The common themes, which vary state by state:
- Imagery should be one input among several, not the sole basis for adverse decisions.
- When imagery is unclear, outdated or low quality, insurers are expected to get more information, often through an on-site inspection.
- Insurers should make sure images are current and correctly interpreted.
- Some states encourage or effectively require sharing the images with the policyholder.
Most of this guidance focuses on underwriting actions like nonrenewals. Maryland's bulletin (25-10) explicitly covers claim denials too. Even where a bulletin only addresses underwriting, the logic carries over: if an aerial report drives your conclusion on a claim, document what you checked on the ground or why an inspection wasn't needed.
This matters more as carriers move inspections off the roof. In October, an executive at a catastrophe adjusting firm told Insurance Business that three carriers had asked in the prior 30 days to avoid field roof inspections, preferring drones or other methods.
Florida: the human-review bill that died
Florida came close to putting the strictest version of this into law. HB 527 and its Senate companion, SB 202, would have barred reducing or denying a claim based solely on an AI or machine learning review. A qualified human would have had to independently analyze the facts and policy terms, check the AI output and reach the conclusion. HB 527 passed the House unanimously in March 2026, then died in the Senate Rules Committee. SB 202 died in Senate Banking and Insurance, according to Repairer Driven News.
The idea isn't dead, though. A unanimous House vote is a strong signal, and similar bills may return in future sessions.
Why this is landing now
Regulators are responding to what's in the market. Verisk's XactAI now offers photo labeling, note summaries, a Line-Item Advisor that suggests line items and an AI Estimator that drafts estimates from plain-language descriptions. Verisk says outputs require human review. Other vendors offer AI-generated mitigation scopes and estimates from 360° capture.
Scrutiny is rising too. The October 2 Warren-Hawley letters to large carriers asked how they use AI in claims and what safeguards they have. In Oklahoma, unsealed documents in private litigation against State Farm reference "touchless claims" plans; State Farm disputes the plaintiffs' characterization. We covered both in "Closed without payment" under scrutiny.
Policyholder attorneys have also started writing about requesting AI outputs and governance documents in discovery. Your notes may be the clearest record of where the tool stopped and your judgment started.
What to put in your file notes
You don't need an essay on every file. A few consistent habits cover most situations.
Name the tool and what it did
"Photo labels generated by [tool], reviewed and corrected." "Initial scope drafted with AI line-item suggestions; reviewed line by line." "Aerial report dated [date] used for roof measurements." One line is enough.
Record your review and any overrides
If you changed an AI suggestion, say so and why. "Removed suggested full tear-off; damage limited to the north slope per field inspection." Overrides backed by observation are evidence of exactly the human review regulators expect.
Tie conclusions to what you observed
For any adverse decision, such as a denial, a partial denial or a repair-instead-of-replace call, the support should be your inspection findings and the policy language, not the tool's output alone.
Handle imagery carefully
Note the capture date of any aerial image. If it predates the loss or is unclear, say what you did about it. If you didn't inspect in person, note why and what else supports the conclusion.
Keep originals
Keep your original, timestamped photos. AI labels and edits are convenient, but the original files are the record. Carriers are also increasingly screening images for manipulation, so original metadata protects you.
Follow the carrier's AI guidelines, and note them
If a carrier or IA firm requires certain tools or review steps, follow them and note that you did. If you're unsure whether a tool is approved for a carrier's files, ask before you use it.
Your own repeatable workflows count too. A scope built from macros and templates you've vetted is easy to explain line by line, which is the point of all this documentation.
What to watch next
- NAIC Fall National Meeting (Nov. 14–17). Watch whether the AI evaluation tool is adopted.
- More states adopting the model bulletin. The count has grown steadily. Check the NAIC map for states where you hold licenses.
- Florida's 2027 session. Watch for a return of the human-review bill.
- Carrier SOP updates. Expect carriers and IA firms to add AI-review language to file standards, especially in Texas.
- Oklahoma litigation. Testimony about automation in claims handling could shape how regulators view these tools.
Status as of November 2026, based on sources dated through October 29, 2026. General information, not legal advice; policy language and state law control.