USPTO Inventorship Guidance Reshapes Carrier AI Patent Filing Strategy

The USPTO's revised inventorship guidance, published in the Federal Register on November 28, 2025 as Document No. 2025-21457, rescinds the February 2024 framework and removes the separate standard for AI-assisted inventions. An AI system is now classified as a tool, analogous to laboratory equipment or a research database.

For carriers filing on machine learning underwriting, pricing, and claims systems, the change removes a prosecution burden that fell hardest on exactly the filings insurers make most.

Key Takeaways

The Pannu factors no longer apply to single-inventor AI cases. The traditional conception test governs uniformly, and the human inventor needs only to have formed a definite and permanent idea of the complete invention.
State Farm holds 326 AI patents since 2014, USAA 218 and Allstate 136, together 77% of all insurance industry AI patents, and P&C insurers account for 89% of insurer AI patents over that period.
Generative AI rose from 4% to 31% of filings between 2014 and October 2025, and generative filings were the hardest cases under the rescinded standard because the boundary between human conception and machine output was the contested question.
Each avoided office action response saves $3,000-$8,000. For a portfolio the size of State Farm's, the guidance removes a whole class of them from pending continuations.
Section 101 is untouched. Recentive Analytics v. Fox Corp. still holds that applying generic machine learning to a new field is not a technological improvement, so the eligibility hurdle is where it was.

What the Rescinded Standard Actually Required

The February 13, 2024 guidance applied the Pannu v. Iolab Corp. factors (155 F.3d 1344, 1351, Fed. Cir. 1998) to AI-assisted inventions. Pannu was designed to test whether multiple human contributors qualify as joint inventors, and it asks for a significant contribution to at least one claim, of non-insignificant quality against the full scope of the invention, amounting to more than explaining the state of the art.

Applied to a single human working with an AI system, that framework asked the applicant to separate their contribution from the machine's as though the machine were a candidate joint inventor. The revised guidance drops that overlay and returns to conception: whether the natural person possessed knowledge of all the limitations of the claimed invention so clearly that only ordinary skill would be needed to reduce it to practice.

Where multiple humans are involved, traditional joint inventorship principles including Pannu still apply, but among the human contributors only. The guidance implements Executive Order 14179 of January 23, 2025 and covers utility, design, and plant patents alike.

The practical effect for insurers was concentrated in prosecution cost and in a category of office action that no longer has a basis. Mayer Brown's analysis still recommends documenting how human contributors selected AI-identified features and integrated them into the claimed solution, but that documentation is now a litigation defence rather than a filing requirement.

The Benefit Lands Where the Portfolios Already Are

The relief is not evenly distributed, because insurer AI patents are not evenly distributed.

Carrier	AI Patents (Since 2014)	Key Filing Areas	Inventorship Guidance Impact
State Farm	326	Claims triage, autonomous vehicle fault, pricing ML	Largest portfolio; most pending continuations affected
USAA	218	Generative AI imagery, agentic AI, property damage	GenAI filings benefit most from simplified standard
Allstate	136	In-vehicle AI, behavior-based pricing, claims automation	Active filing program in telematics AI models
All other insurers	~200 (combined)	Various	Lower volume reduces absolute benefit

Per Insurance Journal's analysis of Evident data, State Farm, USAA and Allstate hold 77% of all insurance industry AI patents, and P&C insurers account for 89% of insurer AI patents filed since 2014. The carriers with the most pending continuations are the carriers with the most filings drafted under the old standard, so the guidance retires a class of office action responses at $3,000-$8,000 each precisely where the pending volume sits.

Generative AI is where the change bites hardest. Filings moved from 4% to 31% of the total between 2014 and October 2025, and those were the applications the rescinded standard handled worst: when a model produces a novel output and the inventor selects and integrates it, the old test demanded a line be drawn exactly where the drafter could least defend one. USAA's filings describing generative systems for clarifying aerial imagery in property damage assessment are the archetype.

What that concentration means for actuarial work is more specific than a general IP observation. The patents in question describe predictive pricing models, loss ratio optimization, claims severity classifiers and telematics-based risk segmentation, which is to say rating methodology. A carrier that patents a machine learning pricing approach constrains competitors to license or design around it, and 77% of that estate sits with three carriers. Industry-wide filing peaked in 2020 and has not fully recovered, so the window for a defensive position is open on a lower base than the numbers above suggest.

The Hurdle That Did Not Move

Inventorship governs who may be named. Section 101 governs whether the invention is patentable subject matter at all, and the November guidance touches only the first.

The Alice/Mayo two-step analysis is unchanged in the federal courts. For an insurance application describing an underwriting algorithm or a risk classification system, that remains the primary obstacle, and the Federal Circuit tightened rather than loosened it in April 2025. Recentive Analytics v. Fox Corp. invalidated AI patents applying generic machine learning to entertainment scheduling, holding that implementing an abstract idea with generic models on conventional infrastructure is not a technological improvement, and that applying established machine learning methods to a different field does not create an eligible invention.

That reasoning maps directly onto a large share of insurer AI filings, which by construction apply established methods to a specific line of business.

The asymmetry is procedural as well as doctrinal. The USPTO's examination posture rejects only where it is more likely than not that a claim is ineligible; a federal court invalidates on clear and convincing evidence. Those are different tests applied at different times to the same patent, and the first one has just become friendlier while the second has not. Venable's December 2025 analysis makes the point that patents granted under current examination may still face invalidity challenges under existing court doctrine.

Greenberg Traurig put the consequence plainly in March 2026: companies treating the changes as an invitation to file broadly without precision may find themselves with a portfolio that looks strong until it is tested. For a carrier whose rating methodology is the asset being protected, the gap between a granted patent and an enforceable one is the whole value of the filing.

USPTO Inventorship Guidance Reshapes Carrier AI Patent Filing Strategy

Key Takeaways

What the Rescinded Standard Actually Required

The Benefit Lands Where the Portfolios Already Are

The Hurdle That Did Not Move

Further Reading

Sources