AI Safety & Assurance Patents: Who Leads, Where the Gaps Are 2026
- 27.2% concentration at the top. The five most active filers account for 100 of the 368 records in scope, but a long tail of single- and few-filing entrants fills out the rest of the ranking.
- Filings grew 46% from 2021 to 2024. Volume rose from 46 records in 2021 to 67 in 2024 — the last year the trend can be read as complete, since publication lag understates 2025 and 2026.
- G06N dominates, but H04L and G16H are thinner. 61.1% of records touch AI-model computing (G06N), while digital transmission (15.8%) and healthcare informatics (7.6%) carry far less claim density.
Filing growth compares 2021 (46 records) with 2024 (67) — 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 368 records in scope (CR5), not by the ranked leaders only.
What this landscape covers
This landscape draws on 368 published records matching claims and descriptions around trustworthy AI, model assurance and AI safety, cross-referenced against core machine-learning implementation terms such as training data, model parameters and neural network inference. The scope runs from 2015 through the August 2026 data cut-off, capturing filings from the earliest wave of governance-oriented AI patents through the current filing year.
Records are drawn from national and PCT filings and grouped by patent family where the ranking counts assignees, so continuation and multi-jurisdiction filing by the same applicant does not inflate the picture. Readers should treat the most recent one to two years as a floor rather than a ceiling: publication typically lags filing by around 18 months, so 2025 and 2026 volumes will keep rising as more applications publish.
Filing trend and technology composition
Two views of the same 368 records: how filing volume has moved year over year, and which technology classes carry the claim density.
Filing trend, 2017–2026
Filings climbed from zero in 2017 to a peak of 102 in 2025, with growth of 46% between 2021 (46 records) and 2024 (67 records) — the most recent span that can be read as complete. 2026 is running at 53 records so far, a partial-year figure that will rise as later publications land.
Technology composition by IPC subclass
G06N (AI-model computing) touches 61.1% of the 368 records, well ahead of G06F general data processing at 32.1%. Because a single record can carry several IPC classes, these shares sum to more than 100% and should be read as claim density per class, not as a partition of the field.
Shares are the percentage of the 368 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 Patent Landscape with Eureka
This page is one run against one query. Ask Eureka your own question about ai safety, evaluation & assurance patent landscape and every answer comes back with the patent numbers behind it.
Try EurekaRepresentative and most-cited filings
Dynamic smart contract security and verification system using capsule networks, autoencoders, and generative adversarial networks
The system pairs an autoencoder that preprocesses smart contract code to reduce noise and surface critical features with a capsule network that captures hierarchical relationships and dependencies in that code. A generative adversarial network generates optimal routing coefficients for the capsule network, and a blockchain-based platform deploys and continuously monitors the resulting contracts.Filed by Leptude, Inc.; published 2026-04-09 as US20260100856A1.
| # | Publication no. | Patent title | Citations |
|---|---|---|---|
| 1 | US20230245651A1 | Enabling user-centered and contextually relevant interaction | 247 |
| 2 | US20180121766A1 | Enhanced human/machine workforce management using reinforcement learning | 77 |
| 3 | US20210117760A1 | Methods and apparatus to obtain well-calibrated uncertainty in deep neural networks | 73 |
| 4 | US20220156614A1 | Behavioral prediction and boundary settings, control and safety assurance of ML & ai systems | 43 |
| 5 | US20190171950A1 | Method and system for auto learning, artificial intelligence (AI) applications development, operationalizatio… | 43 |
| 6 | US20210035021A1 | Systems and methods for monitoring of a machine learning model | 37 |
| 7 | US20250390498A1 | System and method for estimating confidence and implementing metacognitive abilities in artificial intelligen… | 33 |
| 8 | US20260017386A1 | Systems and Methods for Protecting Machine Learning (ML) Units, Artificial Intelligence (AI) Units, Large Lan… | 23 |
| 9 | US20210385135A1 | Action Recommendation Engine (ARE) of a closed-loop Machine Learning (ML) system for controlling a network | 20 |
| 10 | US20260073058A1 | System and method for ai safety red-teaming with policy fuzzing and adversarial prompting | 19 |
Citation counts are drawn from within this searched corpus and favour older filings that have had more time to accumulate citations — read them as a signal of influence on the field, not of current commercial weight.
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 numbers mean for filing strategy
Three findings that shape where a new filing is likely to land cleanly, and where it will compete against dense prior art.
The top of the field is concentrated, not dominant
The five most active assignees account for 100 of the 368 records in scope, and the top ten reach 141 records (38.3%). That leaves close to two-thirds of filings spread across a long tail of companies filing once or twice — a pattern typical of a field still forming its competitive core rather than one already settled by a handful of incumbents.
Growth is real, but recent years understate it further
Filing volume rose from 46 records in 2021 to 67 in 2024, a complete-year comparison that shows genuine acceleration rather than noise. The 2025 peak of 102 and the 2026 partial count of 53 likely continue that trajectory, but because publication lags filing by roughly 18 months, neither year is a reliable read on the field's true current pace yet.
Core AI-model claims are dense; applied verticals are not
G06N (AI-model computing) appears in 61.1% of the 368 records, making core model-level safety and assurance claims the most contested ground. Applied domains carry far less density: healthcare informatics (G16H) sits at 7.6% and image/video recognition (G06V) at 5.4%, suggesting vertical-specific assurance claims remain comparatively open.
Eureka can read the same corpus for gaps instead of for coverage: under-claimed branches adjacent to ai safety, evaluation & assurance patent landscape, with the prior art for and against each one.
Who is filing, and where activity is cooling
The ranked leaders span consultancies, telecom equipment makers and diversified technology companies rather than a single specialist cluster, and recent-year momentum data shows several of the most active historical filers have gone quiet in the latest tracked year.
A single leader sits ahead of a tightly packed group
The leading assignee holds 28 records, comfortably ahead of fifth place at 12 and tenth place at 7. The gap between first and fifth is wide, but from fifth to tenth the field compresses quickly, meaning the mid-table is genuinely contested rather than settled.
Several historically active filers show no latest-year filings
Recent-year momentum data shows multiple assignees that built substantial portfolios earlier in the window recording 0 filings in the latest tracked year, a -100% year-over-year change. Given the 18-month publication lag, this likely reflects unpublished pending applications as much as an actual pullback.
Co-filing is rare and clusters around a few corporate families
Only five co-assignee pairs appear across the dataset, and the strongest links sit within single corporate groups filing jointly across affiliated entities rather than between unrelated companies. Cross-company joint filing is not yet a meaningful pattern in this field.
| Assignee | Recent year | YoY |
|---|---|---|
| International Business Machines Corporation (IBM) | 1 | 0% |
| IQ Consulting Company | 0 | -100% |
| FORT ROBOTICS INC | 0 | -100% |
| Nokia Technologies Oy | 0 | -100% |
| Fujitsu Ltd | 0 | -100% |
| Apple Inc | 0 | — |
| SEEKR TECHNOLOGIES INC | 0 | -100% |
| Nokia Solutions and Networks Oy | 0 | -100% |
Where to take this next
The dataset points to specific next steps depending on whether the goal is freedom-to-operate, portfolio strategy or tracking a competitor.
Check freedom-to-operate against the core cluster
Any filing touching model-level safety or assurance logic sits close to the 61.1% of records classified under G06N. Before drafting claims in that space, map them against the most-cited records to see how broadly the existing claim language already reaches.
Explore the citation network in EurekaWatch the mid-table for consolidation signals
The compression between fifth and tenth place, combined with several leaders showing flat or negative recent-year momentum, suggests the competitive set is still shuffling. Tracking momentum changes quarter over quarter is more informative here than a single snapshot ranking.
Set up assignee tracking in EurekaScope claims toward the thinner verticals
Healthcare informatics, wireless-network safety monitoring and image/video recognition all carry single-digit-to-low-double-digit record shares. These branches are candidates for claims that would not immediately collide with the dense G06N core.
Run a white-space search in EurekaCommon questions about AI safety and assurance patents
This dataset identifies 368 published records matching AI safety, model assurance and trustworthy-AI language combined with core machine-learning implementation terms, covering filings from 2015 through the August 2026 data cut-off. That figure counts published patent documents and applications, not granted patents specifically, and it will keep growing as more recent filings publish. Because publication lags filing by roughly 18 months, the true 2025 and 2026 totals are higher than what is currently visible.
The assignee ranking covers 100 companies counted by record, with the leader holding 28 records ahead of a fifth-place company at 12 and a tenth-place company at 7. The top five combined account for 27.2% of all 368 records in scope, and the top ten reach 38.3%, which means the field has a clear leading group but no single dominant owner of the technology. The remainder of the ranking is a long tail of companies with a handful of filings each.
Filing volume grew 46% between 2021 (46 records) and 2024 (67 records), the most recent span that can be treated as a complete comparison. Headline figures for 2025 (102 records) and 2026 (53 records so far) look like a plateau or decline, but that is an artefact of publication lag rather than an actual slowdown — patent applications typically take about 18 months to publish, so recent years are still filling in. Nothing in this data supports describing the field as cooling.
Computing arrangements based on AI models, IPC class G06N, appear in 61.1% of the 368 records in scope, making it by far the most heavily claimed category. General electric digital data processing (G06F) follows at 32.1%, and digital information transmission (H04L) at 15.8%. Applied verticals such as healthcare informatics (7.6%) and image or video recognition (5.4%) carry much lower claim density, which points to more open space for domain-specific assurance claims.
US20260100856A1, filed by Leptude, Inc. and published 2026-04-09, describes a system for dynamic analysis and verification of smart contracts. It combines an autoencoder for preprocessing contract code, a capsule network for capturing hierarchical relationships in that code, and a generative adversarial network that generates routing coefficients to improve the capsule network's efficiency, all deployed on a blockchain-based monitoring platform. Anyone building smart-contract verification tools that combine these three neural architectures should review its claim scope closely.
Research AI Safety, Evaluation & Assurance Patent Landscape in depth with Eureka
Go past this page: query the whole ai safety, evaluation & assurance 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.