Hallucination Evaluation Patents: Who Leads, Where the Gaps Are 2026
Filing growth compares 2021 (361 records) with 2024 (938) — a three-year span. 2024 is the most recent year we treat as complete: publication lags filing by roughly 18 months, so 2025 onwards are still filling in and any growth rate that ends there would understate the field. Top-5 share is the combined record count of the five largest assignees divided by all 8,105 records in scope (CR5), not by the ranked leaders only.
What the hallucination evaluation patent record actually covers
The search string behind this landscape pulls in two distinct meanings of hallucination: the clinical and pharmacological sense, where hallucination is a side effect being monitored or treated, and the emerging AI-safety sense, where hallucination means a language model generating unsupported output. Records tagged under A61K and A61P dominate the raw count, reflecting decades of pharmaceutical filings that use the word incidentally. The AI-native cluster, concentrated in G06F and G06N, is smaller by volume but newer and still accelerating.
For a reader focused on AI system evaluation, the useful signal sits inside the G06N and G06F subsets rather than the aggregate. The representative record from McKinsey & Company shows where the frontier of claim drafting is heading: scoring the reliability of a language-model response rather than describing the model itself.
Filing trend and technology composition
Filings in scope span 2015 through the mid-2026 data cut-off, with the most recent one to two years still filling in as publications catch up with filing dates.
A field accelerating since 2021
Documented filings ran at 309 in 2017 and climbed to a peak of 954 in 2025. The reliable window for trend reading stops at 2024, where the total reached 938 — up 160% from 361 in 2021. Treat 2025 and 2026 as undercounts rather than a plateau.
Two claim territories, one search term
A61K (56.3% of records) and A61P (38.7%) mark the pharmaceutical-safety territory; G06F (16.8%) and G06N (11.5%) mark the AI-evaluation territory. C07D, C12N and C07C track the chemistry side of the pharma cluster, while G01N covers analytical testing methods that increasingly appear in both worlds.
Shares are the percentage of the 8,105 records in scope. A patent can carry several IPC classes, so the shares add up to more than 100%.
Go deeper on AI Safety, Evaluation & Assurance: Hallucination Evaluation Patent Landscape with Eureka
This page is one run against one query. Ask Eureka your own question about ai safety, evaluation & assurance: hallucination evaluation patent landscape and every answer comes back with the patent numbers behind it.
Try EurekaA representative filing at the AI-evaluation frontier
Contextualized output reliability evaluation for language models
A method of evaluating the reliability of a response generated by a language model is described herein. The method includes obtaining a query provided to a language model and a response to the query generated by the language model, and applying hallucination evaluation techniques to one or more of the query and the response to generate hallucination scores. The method also includes combining the hallucination scores to obtain a confidence score, and providing an indication of the confidence score for display on a user interface, where the confidence score indicates the reliability of the response.Filed by McKinsey & Company, this April 2026 application is one of the clearest examples of the AI-native branch of this landscape: it claims a scoring pipeline around a model's output rather than the model itself.
| # | Publication no. | Patent title | Citations |
|---|---|---|---|
| 1 | WO1998027230A1 | Methods and compositions for polypeptide engineering | 1,316 |
| 2 | US6586182B1 | Methods and compositions for polypeptide engineering | 575 |
| 3 | US20060161218A1 | Systems and methods for treating traumatic brain injury | 496 |
| 4 | WO2016110804A1 | Mobile wearable monitoring systems | 466 |
| 5 | US20210169417A1 | Mobile wearable monitoring systems | 453 |
| 6 | US6335160B1 | Methods and compositions for polypeptide engineering | 443 |
| 7 | US6303344B1 | Methods and compositions for polypeptide engineering | 431 |
| 8 | US20090312817A1 | Systems and methods for altering brain and body functions and for treating conditions and diseases of the same | 410 |
| 9 | US20170365101A1 | Augmented reality display system for evaluation and modification of neurological conditions, including visual… | 403 |
| 10 | US6319713B1 | Methods and compositions for polypeptide engineering | 397 |
Citation counts favour older filings simply because they have had longer to accumulate citations inside this corpus — read them as a signal of influence on the field, not of current commercial relevance.
Each row carries its publication number; clicking a row searches Eureka by that number.
Put your own technology through the same analysis
Eureka on the web
When you want the answer in the next five minutes.
The agent works the prompt against patents and technical literature, citing every source.
Run your analysis now →MCP server & REST API
When it has to run inside your own pipeline.
Patent search, landscape analysis and assignee resolution as MCP tools. Drop them into any agent framework, or call REST directly.
Browse MCP servers →What the concentration and technology figures mean for filing decisions
Three figures from this dataset matter more than the raw record count when deciding where to file or where to look for prior art.
No single company controls this space
The leading assignee holds 170 records against a field of 8,105, and the top 5 combined reach only 8.8% of all records in scope. That is a genuinely fragmented landscape — useful context before assuming any one filer's claims are unavoidable.
The AI-evaluation branch is filing faster than it is being cited
The 2021-2024 window is the most reliable growth signal in this dataset because publication lag has fully cleared it. Filings nearly tripled over three years, well ahead of what the smaller, older citation counts on record would suggest.
AI-native claims are still a minority of the corpus
G06N and G06F together cover the language-model side of hallucination evaluation, but they sit well behind the pharmaceutical A61K and A61P clusters by raw volume. That gap is a function of the search term's dual meaning, not of the AI branch being small in absolute terms.
Eureka can read the same corpus for gaps instead of for coverage: under-claimed branches adjacent to ai safety, evaluation & assurance: hallucination evaluation patent landscape, with the prior art for and against each one.
| Assignee | Co-assignee | Shared families |
|---|---|---|
| Ionis Pharmaceuticals, Inc. | Genzyme Corporation | 10 |
| Flamel Ireland | Richter Gedeon Nyrt. | 8 |
| H. Lundbeck A/S | LI GUIYING | 4 |
| H. Lundbeck A/S | DOLLER DARIO | 4 |
| H. Lundbeck A/S | MA GIL | 3 |
| H. Lundbeck A/S | GRENON MICHEL | 3 |
| Newron Pharmaceuticals S.p.A. | THALER FLORIAN | 3 |
| Newron Pharmaceuticals S.p.A. | SALVATI PATRICIA | 3 |
The United States (2,795 records) leads receiving offices by a wide margin, followed by Europe (1,009), WIPO/PCT filings (894), Australia (491), Israel (403) and India (388) — a spread that suggests applicants are pursuing broad multi-jurisdiction protection rather than concentrating in one office.
Where to take this analysis
This landscape is a starting point for freedom-to-operate and whitespace work, not a substitute for it.
Trace the AI-native subset in detail
Isolate the G06N and G06F records from the pharmaceutical noise and run a family-level review of claim scope around confidence scoring, output verification and reliability metrics.
Open Eureka to filter this datasetWatch the fragmented leaderboard
With the top 5 assignees holding under 9% of records combined, new entrants still have room to establish a defensible position — track new filings from adjacent AI and enterprise software players.
Set up assignee monitoring in EurekaModel the 2025-2026 undercount
Before concluding the field is slowing, adjust for the roughly 18-month publication lag and revisit the trend once later filings resolve.
Explore filing trends in EurekaCommon questions about the hallucination evaluation patent landscape
This dataset contains 8,105 published records matching the search terms across 2015 through mid-2026. That figure spans two distinct uses of the term: pharmaceutical filings where hallucination is a monitored side effect, and AI-safety filings where hallucination means unsupported model output. Because the two meanings share the same keyword, the raw total overstates the size of the AI-native branch specifically, which is better read through the G06N and G06F subclass counts.
No. The leading assignee in this dataset holds 170 records out of 8,105, and the top 5 assignees combined account for just 8.8% of all records in scope. That is a low concentration for a patent landscape, meaning there is no single blocking filer and the field remains open to new entrants who file well-drafted claims. The tenth-ranked assignee holds 96 records, showing a long tail rather than a steep drop-off after the leader.
Yes, through the last fully reliable year. Filings rose from 361 in 2021 to 938 in 2024, a 160% increase, and the field peaked at 954 filings in 2025 on current publication data. Because patent publication typically lags filing by around 18 months, the apparent slowdown in 2025 and 2026 figures reflects incomplete data rather than an actual drop in filing activity.
The dataset splits across pharmaceutical and AI-computing classifications. A61K (medicinal preparations) covers 56.3% of records and A61P (therapeutic activity) covers 38.7%, reflecting the clinical use of the term. G06F (electric digital data processing) covers 16.8% and G06N (AI-model computing) covers 11.5%, marking the branch relevant to language-model evaluation. A single record can carry multiple classes, so these figures do not sum to 100%.
A representative example is US20260093931A1, filed by McKinsey & Company and published April 2026, which claims a method for scoring the reliability of a language model's response by generating hallucination scores and combining them into a confidence score shown on a user interface. This reflects a broader shift in recent AI-native filings toward evaluating model output after generation, rather than claiming the underlying model architecture itself. Filings of this type sit inside the smaller G06N/G06F cluster of the overall dataset.
Research AI Safety, Evaluation & Assurance: Hallucination Evaluation Patent Landscape in depth with Eureka
Go past this page: query the whole ai safety, evaluation & assurance: hallucination evaluation patent landscape corpus yourself, in your own scope.
Every answer comes back with patent numbers you can open.
Disclaimer. This page is generated from Patsnap Eureka data drawn from a limited snapshot of global patent and scientific-literature records, and is provided for general information and reference only.
Patent data carries inherent limitations: recent filings (typically the most recent 18–24 months) are under-counted due to standard publication lag; counts may be reported at either a patent-family or a patent-record basis and are not always directly comparable; classification, applicant-name, and citation data may contain errors, duplicates, or omissions; and the underlying search query defines and constrains the scope shown. As a result, the analysis may be incomplete or inaccurate and may not reflect the full technology landscape.
Nothing on this page constitutes an exhaustive prior-art, novelty, freedom-to-operate, or validity search, nor does it constitute legal, financial, investment, or professional advice, and it should not be relied upon as such. Any patent, commercial, or strategic decision should be verified independently and reviewed with qualified patent, legal, and domain professionals. Patsnap makes no warranties, express or implied, as to the accuracy, completeness, or fitness for any particular purpose of the information presented.
Machine translation. Assignee and organisation names originally recorded in Chinese, Japanese or Korean have been rendered into English by an AI translation step so that the tables stay readable. These renderings are best-effort and may not match a company’s registered English name; the original name is what the underlying patent record carries, and it is what any Eureka query launched from this page uses.