USAA Took 19 Patent Grants in July; Filing Leader State Farm Took None

The grant record, not the filing count, shows which insurer AI survives a patent examiner, and July 2026 inverted the league table: USAA took 19 US utility grants across the July 7, July 14 and July 21 issue dates and Allstate took three, while State Farm, whose 326 AI-related filings since 2014 lead the industry, took none (USPTO Official Gazette, July 2026; Insurance Journal, December 2025).

USAA utility patents granted across July 2026's three issue dates

Allstate grants in the same three Official Gazette issues

July grants listed under the State Farm name, against an industry-leading filing stock

actuary.info has covered the filing side of this race, including the December 2025 finding that State Farm, USAA and Allstate hold 77% of insurer AI patents tracked since 2014 (Insurance Journal, December 2025). Filings measure intent. A grant is different: it is a set of claims that survived prior-art search and subject-matter review, and that competitors and vendors must now design around for up to two decades. Reading the three July issue dates patent by patent shows what actually cleared that bar. The answer concentrates in claims operations: triage that opens the file before the policyholder calls, injury severity scored from cabin video, and a ruling on which policy was on the hook at a rideshare crash.

The July Grant Sheet: 22 Patents to Two Carriers

US utility patents issue on Tuesdays, and July 2026 has had three issue dates so far, with a fourth arriving July 28. The patentee indexes of the USPTO Official Gazette for those three issues list 19 grants to United Services Automobile Association: seven on July 7, seven on July 14 and five on July 21. The same indexes show three grants to Allstate and none to State Farm Mutual Automobile Insurance Company (USPTO Official Gazette, July 2026). Swiss Re is absent from all three weekly indexes as well. Subsidiary grants under other names would not surface in those entries, but the top-line contrast stands. The carrier with the industry's largest AI filing stock put nothing on the grant ledger in July, while USAA averaged better than six grants per issue date.

Not all 19 USAA grants are AI patents. Security and infrastructure carry a large share of the sheet: two network anomaly-detection grants classified in H04L 63/1425, a blockchain grant in H04L 9/50, and a scatter of interface, payments and data-management classes. Two others, a vehicle-theft response system and a grant in the airspace-management class G08G 5/32, mark the operational edges of auto and property servicing. The four grants with the most direct reach into P&C pricing and claims work sit in the analytics stack, and each is worth reading at the claim level.

Patent	Carrier	Title	Issued	Application filed	Earliest date in family
US 12,682,402 B1	USAA	Intelligent methods of inspection for property and casualty insurance claims	July 14, 2026	October 1, 2024	March 8, 2013
US 12,682,662 B1	USAA	Systems and methods for determining likelihood of traffic incident information	July 14, 2026	June 6, 2023	October 31, 2016 (provisional)
US 12,680,947 B1	USAA	Systems and methods for detecting properties relating to building components	July 14, 2026	March 14, 2024	February 19, 2019 (provisional)
US 12,675,828 B2	Allstate	Ride for hire	July 7, 2026	July 23, 2024	July 23, 2024 (no earlier filing listed)
US 12,688,779 B2	Allstate	Roadside assistance system	July 21, 2026	March 28, 2024	May 16, 2019

US 12,682,402: A Loss Report Before the Policyholder Calls

The most consequential grant of the month is US 12,682,402 B1, issued to USAA on July 14, 2026 in the insurance data-processing class G06Q 40/08. Claim 1 recites a pipeline any CAT actuary will recognize: identify a bounded geographical area affected by a weather event, build a customer density profile inside it, and collect sensor data from the area. A severity code estimating the level of property losses follows from those three inputs, and the code drives a prediction of resource utilization. The claim's last two steps are where the fence goes up. The system generates loss reports for selected customers "prior to receiving customer-generated loss reports from the selected customers" (US 12,682,402 B1, July 14, 2026), then deploys the predicted resources against them.

In plain terms, USAA now holds an issued claim on inverting first notice of loss for catastrophe claims: the carrier opens the file, codes its severity and stages the adjuster capacity before the insured picks up the phone. Nothing in claim 1 recites a neural network or any specific model; the limitations are functional, a computer system comprising one or more processors performing the steps. That breadth cuts both ways. It means the claim does not depend on any particular imagery or ML technique. A vendor platform that footprints a hurricane from weather and sensor feeds, scores expected severity against a customer file and pre-stages field resources is running the same architecture the claim describes. Whether any given product practices every element is a question for counsel, but the elements are generic enough that the answer is not obviously no. The same breadth is exactly what post-Alice eligibility review targets in G06Q art, a tension the site examined when the USPTO put GenAI tooling into the examining corps this month. This family has been running that gauntlet since before Alice was decided: the July grant is the fifth application in a continuation chain running back to a March 8, 2013 filing, a 13.4-year priority-to-grant span (USPTO Official Gazette, July 2026).

US 12,682,662: Injury Severity From Two Frames of In-Cabin Video

Issued the same day, US 12,682,662 B1 claims a camera and controller that, on an indication that a traffic incident has occurred, capture two images of a vehicle occupant and identify a shared biomechanical point across the frames. The controller computes the distance, speed or acceleration of that point between frames, converts the movement into the force the occupant absorbed, and scores a severity of injury from the force. The classification list includes probabilistic pattern classification (G06F 18/2415) alongside vehicle-interior scene recognition (G06V 20/59), and the family runs back through four applications to an October 31, 2016 provisional (USPTO Official Gazette, July 2026).

For a casualty actuary, this is bodily-injury triage moved to the moment of impact. A severity estimate computed from occupant kinematics, before any medical bill exists, is an early reserving signal on the injury coverages, a routing input for nurse-line and SIU referral, and a check against later-presented soft-tissue claims whose alleged mechanism does not match the measured force. The claim-scope seam sits in the force step: claim 1 requires identifying a shared biomechanical point and calculating force from its movement. An end-to-end deep-learning model that maps raw crash video straight to an injury score, without an explicit force computation, arguably walks around the independent claim. Telematics and dashcam vendors extending from driving behavior into occupant monitoring now have to make precisely that design choice. The gazette entry also notes a terminal disclaimer, the standard marker of a family dense enough that USAA is stacking near-duplicate claim sets around the same invention, which is how a carrier builds a thicket rather than a single fence.

US 12,680,947: Material Type From Reflected Sound

The third July 14 grant in the featured set, US 12,680,947 B1, lists 20 named inventors and claims a sensor deployed near a building that reads sound or vibrations reflected off building components. Combined with a room layout including size and wall positions, the readings determine the material type of those components. The family traces to a February 19, 2019 provisional and has already produced three earlier patents, including US 11,105,735, granted August 31, 2021 (Google Patents, July 2026). That parent's specification describes matching detected sound signatures against a database of known floor types rather than running a trained model.

The precise read: this family is sensing plus database lookup rather than machine learning, and the claims protect a data-capture channel rather than any model that consumes its output. Replacement-cost estimation, ITV auditing and underwriting inspection all depend on knowing what a structure is made of, and the current market answer is a photo app plus a human or a vision model. An acoustic method that identifies materials without an adjuster on site is a different capture channel entirely. Vendors selling photo-based inspection tools do not obviously touch these claims, which makes the patent less a litigation weapon than an option on a sensing modality nobody else can commercialize without a license. Twenty inventors on one family measures how much internal engineering USAA has put behind that option.

Allstate's July Grants: Coverage Attribution at the Moment of Impact

Allstate's US 12,675,828 B2, titled Ride for Hire, issued July 7, 2026 and classified in G06Q 40/08 alongside the transportation-services class G06Q 50/47, automates the question that decides which insurer pays a rideshare loss. The claimed apparatus detects an accident from telematics data, queries the driver's ride-hailing app for interaction metadata with timestamps, and determines which period of use of the app the accident occurred in. It then notifies the driver's device and alters the operational state of the app. Period of use is the load-bearing phrase: the personal auto policy, the TNC's contingent coverage and the TNC's primary commercial cover attach to different periods of the rideshare cycle, and the boundary dispute between them is a recurring coverage and subrogation fight. An issued claim on resolving that boundary automatically, at the moment of loss, from app metadata rather than driver testimony, sits directly on the loss-allocation seam between two books of business. The application moved from a July 23, 2024 filing to grant in just under 24 months, and the gazette entry lists no earlier family filing (USPTO Official Gazette, July 2026).

The other analytic Allstate grant, US 12,688,779 B2, issued July 21, 2026, is a roadside-assistance dispatch system that ranks service providers on historical performance and equipment-adjusted completion timing, routes the request down the ranked list and updates the rankings from each response. As claims automation it is modest, an optimization loop on vendor dispatch, but ranked dispatch against measured provider performance is the pattern carriers now apply to appraisal networks and contractor panels, and its costs land in LAE. Allstate's third July grant, in the emergency-communications class H04W 4/90, rounds out a month in which every Allstate issuance touched the accident-response chain (USPTO Official Gazette, July 2026).

Pipeline Math: 2013 to 2019 Inventions, 2026 Grants

Every featured USAA grant this month is a continuation, and the chains are long: the CAT triage family runs from March 2013, the occupant-kinematics family from October 2016, the building-components family from February 2019. The final continuation link moved quickly in each case, between 21 and 37 months from application to grant, but the inventions themselves predate the current AI cycle by half a decade or more. July's grant sheet is a map of USAA's 2013-to-2019 engineering agenda, refiled and re-scoped ever since. That is the strategic function of continuation practice: keeping a family pending lets the carrier redraft claims year after year to track where the vendor market is actually going. The claims that finally issue are aimed at the products competitors can buy today, not the prototype of 2013.

The same math sets expectations for the current filing wave. Evident, the AI benchmarking firm whose tracker underlies the December concentration reporting, counted generative AI rising from 4% to 31% of insurer AI patent filings between 2014 and October 2025 (Insurance Journal, December 2025). Only three insurers pursue agentic AI patents at all (Insurance Business America, December 2025). None of July's featured grants recites a generative model, and on observed pendency none should have. The GenAI filings of 2023 through 2025 are still in prosecution and will surface as grants roughly from 2027 onward, with the agentic claims covered in USAA's agentic filing lead behind them. The corollary is the State Farm zero. A 326-filing stock with no July grant flow says less about the stock than about conversion timing: grant flow is lumpy, and timing rather than filing volume decides when a moat exists. USAA's continuation practice produces steady flow. State Farm's stock has to convert eventually, and when it does, the grants will land in a market two years ahead of the claims as drafted.

Concentration, Vendor Dependence and the NAIC Overlay

The December data frames why a single month of grants is worth this much attention. Three carriers hold 77% of insurer AI patents tracked since 2014, on filing counts of 326 for State Farm, 218 for USAA and 136 for Allstate (Insurance Journal, December 2025). The pipe behind them is thin and thinning: 30 major insurers across North America and Europe filed just 166 AI patents from January 2023 through late 2025, running about 30% below the 2020 filing peak, and 89% of the filings sit with P&C carriers (Insurance Business America, December 2025).

For the several hundred US carriers outside that group, each issued claim narrows the build path. A mid-market carrier can still assemble pre-FNOL CAT triage from weather data, aerial imagery and a customer file, but its counsel now reviews that architecture against an issued USAA claim rather than a pending application. The safe harbor is a design-around or a license. In practice the pressure lands on vendors: carriers that buy their CAT triage, injury scoring or dispatch optimization inherit whatever freedom-to-operate exposure their vendor carries, which makes patent indemnification language in AI vendor contracts a live underwriting-of-the-vendor question. That diligence is converging with the regulatory track the site has followed on the NAIC's third-party data and model vendor framework. The same registry-and-attestation workflow that documents a vendor's model governance is the natural place to record whose claims the vendor's product may be practicing. Concentrated grants plus governed vendor dependence is a strong-form version of build versus buy, in which the three carriers that build are also quietly setting the licensing terms of everyone else's buy.

Reading the July Grants From the Actuary's Chair

For reserving actuaries, the two July 14 claims patents describe inputs that will change early development patterns wherever they, or products like them, deploy. A severity code assigned before FNOL and an injury score computed at impact both pull information forward in the claim lifecycle, which compresses the reporting lag embedded in current-triangle assumptions. An actuary watching a book adopt pre-FNOL triage should expect the first diagonal to strengthen and should resist reading that strengthening as deterioration. For pricing, Allstate's period-of-use attribution sits on the boundary between personal and commercial auto experience: automated, metadata-based allocation of rideshare losses will move premium and loss between those books, and rate filings on either side inherit the shift.

The watch items for the rest of 2026 are specific. First, whether the USAA families featured here continue again; a fresh continuation filed off the CAT triage family would signal the 2013 chain is still being aimed at new products. Second, whether State Farm's grant flow resumes at the July 28 or August issue dates; its filing stock is too large for a durable zero. Third, how examiner-side GenAI tooling and the December 2025 eligibility guidance shape conversion of the 2023-2025 GenAI filing wave. The filing league table is two years stale by construction. The gazette's Tuesday grant sheet is the current score, and this month it read USAA 19, Allstate 3, State Farm 0.