Hartford Patents an Outlier Engine for Underwriting and Fraud

Hartford Fire Insurance Company's US Patent 12,700,040 B2, granted August 4, 2026, scores any data point by how far its attribute grouping sits from the norm across correlated data sources. It then feeds that outlier back into the model as a fresh input, and routes the result to an underwriting decision, a fraud investigation, or a claims outcome (USPTO Official Gazette, August 2026).

Trade coverage of carrier patents defaults to the fraud-detection frame, because that is the plainest use case Hartford names in the grant. The more consequential reading sits one layer down.

Executive Summary

An algorithm that flags an attribute grouping as anomalous, and then lets that flag influence an underwriting or fraud-routing decision, is performing risk selection, the same function a filed rating variable performs, without going through the actuarial justification and disparate-impact testing a named variable has to clear before a state regulator will let it touch price or acceptance.

The regulatory frameworks now in force, New York's Circular Letter No. 7 and the NAIC's AI Model Bulletin, were built for named variables with fixed definitions a regulator can test. Hartford's outlier score has neither: "typical" is whatever the model currently computes it to be, and a feedback loop that reinjects flagged outliers means that definition shifts over time.

Reserving actuaries have a separate, concrete concern. A routing engine that changes the rate at which claims get pulled into special investigations changes when and how paid losses emerge, and that shift is indistinguishable from adverse development inside a standard loss triangle.

Set against the Allstate and State Farm underwriting patents granted the same month, both of which expose named thresholds or traceable code changes an auditor can inspect, Hartford's design marks a distinct step from explicit, auditable rules toward inferred structure that resists exactly the documentation regulators now expect.

Patent Details

Patent Number	U.S. 12,700,040 B2
Granted	August 4, 2026 (Official Gazette week 31)
Assignee	Hartford Fire Insurance Company
Inventors	Arthur Paul Drennan III; Tracey Ellen Steger
Classification	CPC G06Q 40/08 (insurance subclass of finance and business methods)
Predecessor	U.S. 11,244,401 B2, "Outlier System for Grouping of Characteristics" (same inventors)
Downstream routing	Underwriting decision, special-investigations referral, or claims outcome

The Mechanism: Grouping, Separation, and a Feedback Loop

The new patent's lineage runs back to an earlier grant from the same inventor pair, Arthur Paul Drennan III and Tracey Ellen Steger, both Hartford data scientists whose names appear on a run of the carrier's analytics patents. Their prior US Patent 11,244,401 B2 describes a system that receives data from multiple external and internal sources, correlates selected parameters including derived characteristics, and models the data by identifying attributes associated with particular groups based on the groupings' relative separation from typical groupings when spatially graphed (Google Patents, USPTO).

The August 2026 grant extends that architecture with a loop. Once the system flags a grouping as an outlier, it reinjects the flagged data as a new input source, so an anomaly the model surfaces in one pass becomes part of what the next pass correlates against. The output then routes to one of three downstream paths: an underwriting decision, a special-investigations referral, or a claims outcome (USPTO, August 2026).

Nothing in the Mechanism Requires a Fraud Signal

Both patents sit under Cooperative Patent Classification G06Q 40/08, the insurance subclass of the finance and business-methods category, rather than a fraud-specific classification, and that placement matches what the claim language actually does. Separation from a typical grouping is a general statistical property; the routing step decides what a given outlier means for a given policy or claim.

A homeowners application whose attribute grouping sits far from the norm can be routed to underwriting the same way a claim with an unusual combination of loss-type, provider, and timing characteristics can be routed to a fraud investigator. The patent is a segmentation engine wearing an anomaly-detection label, and segmentation is exactly the tool a pricing or underwriting actuary already knows how to regulate, just not usually one that discovers its own groups without being told which attributes to watch.

The Problem: Risk Selection Without a Filed Variable

New York's Department of Financial Services drew the clearest line state regulators have drawn on this question. Circular Letter No. 7 (2024) requires insurers using external consumer data or AI systems in underwriting and pricing to "evaluate the extent to which such" data sources "are correlated with status in any protected classes" that could produce discriminatory harm, and to run a three-step test before deployment and on a regular cadence afterward:

• Check for disproportionate adverse effects on protected classes;
• Determine whether a legitimate business reason explains any disparity found;
• Search for a less discriminatory alternative before relying on the variable at all.

The letter also imposes an explainability obligation: insurers must "explain, at all times, how the insurer's AIS operates," and a consumer denied coverage is entitled to a detailed explanation of the information behind that decision within 15 days, with no exception for proprietary vendor logic (NY DFS, 2024).

That framework was built for a named variable: a credit-based insurance score, a territory factor, a motor-vehicle-report tier, something a regulator can point to and test. Hartford's outlier score is not that. It is the output of an unsupervised process that decides for itself which attributes, in whatever combination the current data pull happens to correlate, define a "typical" grouping, and then measures distance from that self-defined center.

There is no single variable to hand a regulator, because the variable is whatever the model currently treats as normal, and the feedback loop means that definition can shift as flagged outliers get reinjected as new inputs. A carrier can run disparate-impact testing on the final routing decision, but testing the decision's output is a weaker check than testing the input variable itself. The input level is where the NAIC's Model Bulletin on the Use of Artificial Intelligence Systems by Insurers expects governance to operate, and the bulletin, adopted December 4, 2023, names underwriting and claims explicitly among the lifecycle stages its documentation and testing expectations cover.

The Exposure Is Concrete for Hartford Specifically

More than 20 U.S. jurisdictions had adopted the NAIC bulletin by mid-2026, including Connecticut, Hartford's own domiciliary state, which issued its version, Bulletin No. MC-25, on February 26, 2024 (NAIC AI Bulletin tracker, July 2026). A carrier's home regulator is typically its most frequent examiner, so the exact documentation standard the outlier patent's structure resists applies first to the department that knows Hartford's filings best.

The industry has already seen what an algorithmic factor without a clean paper trail costs in litigation exposure: State Farm is defending a race-discrimination suit over claims-handling algorithms that plaintiffs argue functioned as an unfiled proxy for protected-class status (Bloomberg Law, 2026). An outlier engine that was never filed as a rating variable, because its designers can accurately say it is not one, does not avoid that exposure. It just moves the fight from a rate hearing to a market-conduct exam or a courtroom.

Three Carrier Patents, One Month, Three Levels of Auditability

Set next to the two other carrier pricing-and-underwriting patents granted this same August, Hartford's design marks a distinct step away from explicit, auditable rules.

Patent	What it automates	What an auditor can inspect
Allstate, US 12,711,555 (August 18, 2026)	Scores a mitigation-adjusted alternative premium against the filed rating plan; auto-rejects when the two diverge past a set threshold, no underwriter required by Claim 1	The threshold, the comparison, and both premium models are named, fixed structures readable off the claim language
State Farm, US 12,694,432 (July 28, 2026)	An LLM flags a specific pricing error, writes the code fix, tests it in simulation, and deploys without human sign-off required by the core claim	Aggressive in what it automates, but traceable to a specific, identifiable error and a specific code change
Hartford, US 12,700,040 (August 4, 2026)	Scores distance from a self-defined "typical" grouping and routes to underwriting, SIU, or claims; flagged outliers feed back into the next pass	No fixed threshold, comparison, or variable; "typical" updates as the loop runs

An auditor can request Allstate's threshold or read State Farm's code diff. An auditor asking Hartford what made a given grouping abnormal is asking a question the system itself may answer differently depending on when it is asked.

Actuarial Implications

Reserving: Fraud Triage Changes Its Own Loss Development

A routing decision that sends a claim to special investigations rather than ordinary adjustment changes when and how that claim's paid losses and allocated loss-adjustment expense emerge. A claim held for SIU review pays out later than one adjusted on the standard track, often substantially later, and the investigation itself adds an expense layer an ordinary claim never carries.

Loss-development factors are built on an implicit assumption that the mix of claims flowing through a given accident period develops in a reasonably stable pattern from one valuation to the next. If an outlier-routing engine changes the rate at which claims get flagged, whether because the model retrains, the feedback loop's reinjected outliers shift what counts as typical, or Hartford simply tunes the separation threshold, the SIU referral rate itself becomes a variable the reserving actuary has to monitor, the same way a change in claims-handling philosophy has to be flagged to whoever is booking reserves.

Development that looks like adverse emergence in a later valuation could just as easily be a change in how many claims get pulled onto the slower investigative track before reaching the same eventual disposition. That distinction matters enormously for reserve adequacy, and it is invisible in the triangle itself.

Governance: Two Uses, Two Very Different Obligation Levels

Hartford's grant names both a data-quality use and a decisioning use for the same outlier-scoring mechanism, and the governance obligations attached to each are not close to equivalent.

Using an outlier flag to catch a malformed or duplicated record before it enters an already-approved rating algorithm is a data-hygiene function, closer in kind to validation a carrier's data warehouse already runs. It does not touch the filed rating plan's inputs or outputs.

Using the same flag as an input that changes whether a policy is accepted, at what tier, or whether a claim gets pulled into investigation is a decisioning function that produces exactly the "Adverse Consumer Outcome" the NAIC bulletin defines: any decision "subject to insurance regulatory standards enforced by the Department that adversely impacts the consumer in a manner that violates those standards" (NAIC Model Bulletin, December 2023).

A claim set broad enough to cover both uses in one specification is standard drafting practice, a pattern actuary.info's coverage of the USPTO's 2026 Section 101 guidance shift found examiners increasingly willing to grant across the finance and insurance classification. The governance question is not whether Hartford can build a mechanism that spans both uses; the patent proves it already has. It is whether the compliance function, at Hartford or any carrier licensing a comparable design, tracks which decisions the outlier score is quietly touching once it leaves the data-cleaning pipeline.

The Tension With Hartford's Own Transparency Record

Hartford published a voluntary Algorithmic Impact Assessment in February 2026, the first among top-20 P&C carriers, disclosing bias audits for named variables including ZIP code, age, and property type across its production AI models (actuary.info's coverage of Hartford's Algorithmic Impact Assessment, May 2026).

That disclosure works because ZIP code, age, and property type are named variables with fixed definitions an auditor can test for protected-class correlation year over year. An unsupervised outlier score has no equivalent fixed definition to publish. The carrier that set the industry's transparency bar for named variables now also holds a patent on a mechanism that, by design, does not have one to disclose.

The volume running through that governance seam is growing. Hartford's Business Insurance segment posted a 5% premium increase and an 89.3 underlying combined ratio in the second quarter of 2026, while Personal Insurance improved to an 86.3 underlying combined ratio despite a 7% decline in written premium, results management attributed in part to automation and AI-enabled underwriting capabilities (Hartford Q2 2026 earnings call, August 2026). Growth built partly on AI-assisted underwriting raises the volume of decisions running through whatever governance process Hartford applies, at the same pace the volume of policies affected by it grows.

What the Filing Question Comes Down To

The patent itself settles nothing about how Hartford has configured the system in production; a granted claim describes what the company can build, not what compliance controls sit around a specific deployment.

But the mechanism as claimed creates a structural mismatch with the tools state regulators currently use to govern AI-assisted underwriting, tools built around named variables, fixed thresholds, and testable definitions of "typical." An outlier score that updates its own definition of normal every time the feedback loop reinjects a flagged data point is harder to file, harder to test for proxy discrimination under a framework like New York's, and harder to explain to a denied applicant inside a 15-day window than the accept-reject engines Allstate and State Farm patented in the same month.

Reserving actuaries watching Hartford's development triangles and pricing actuaries reviewing its rate filings both have the same practical question to ask before the next valuation: whether the SIU referral rate and the underwriting acceptance rate have stayed stable enough, quarter to quarter, that neither one is quietly doing the work of a rating variable that was never filed.