Get the newsletter
Samsung · Networking · Filed Jun 11, 2025 · USPTO published Nov 19, 2026 · This week's release

Samsung Patents a Filing Cabinet for the AI Models Inside Mobile Networks

Before AI models can run across cell networks, someone has to keep track of them. Samsung's patent application describes a registry that gives each model an ID, a description card and a place on the shelf.

The steps needed to build, check, and run an artificial intelligence model before rolling it out across a mobile network. Drawing from patent filing US 2026/0370563 A1.
The steps needed to build, check, and run an artificial intelligence model before rolling it out across a mobile network.All 12 drawings ↓
Publication number
US 2026/0370563 A1(US20260370563A1)
Applicant
Samsung Electronics Co., Ltd.
Filed · Published
Jun 11, 2025 · Nov 19, 2026
Inventors
Chadi KHIRALLAH, Ezeddin Al HAKIM, Mahmoud WATFA
Status when we published
Waiting for an examiner
Document
15 claims
US classification
709/221
Examiner
CHOU, ALAN S (Art Unit 2451)
Parent application
National Stage Entry of PCT/KR2023/019572 (filed Nov 30, 2023)(PCTKR2023019572)

What Samsung's AI model registry actually does

Phones and cell networks are starting to run AI models for jobs like managing connections and tracking position, and every time one of those models gets shipped back and forth, it costs signal traffic, battery and computing power. On top of that, the filing lists open questions: how to save a trained model, what to store with it, and how to find the best one later.

Samsung's answer is a record keeper. A phone sends in a model along with a description card: what it's for, how big it is, how accurate it tested. The keeper assigns an ID number, files everything away, and tells the phone it worked. Later, a device can ask for the best match, such as a model for handling devices on the move that scored above 95 percent accuracy.

You would never see any of it. It would sit in the network's plumbing, helping your phone and its network agree on which AI model to use.

From the filing · CLAIM 1
receiving, from a user equipment (UE), a request to register an AI/ML model or functionality, the request including metadata related to the AI/ML model or functionality; assigning an identifier (ID) to the AI/ML model or functionality; and storing information on the AI/ML model or functionality and the related metadata.

Translation: The system takes in a new AI model from a phone, gives it an ID tag, and saves its details.

How a phone registers an AI model with the network

The patent describes a new network component, called a Model Learning Controller in the description, that acts as the record keeper for AI/ML models (machine-learning programs trained to make predictions). The claims cover its basic loop: it receives a registration request from a user equipment (the phone or other device), assigns the model an identifier, and stores the model's information along with the metadata (descriptive details) that came with the request.

The claimed metadata can include:

  • What the model is used for, plus its inputs and outputs
  • Accuracy and other performance numbers
  • Size, including the number of layers and parameters (the adjustable numbers inside a model)
  • Whether it was trained or runs on the device side or the network side, online or offline

IDs can be global or local. The stored record can include the model's architecture, its weights (the learned numbers), how long it stays valid, and when and where it was registered.

Dependent claims add two more steps. A device can send an update message to change a registered model, and it can send a selection request, which the manager answers by identifying a matching registered model. The description also covers subscribing to changes and configuring models, plus options like averaging the weights of several stored models into one. Those extras are described but not claimed. The manager can live almost anywhere: a base station, a core network function, a server, or even a phone.

What a shared AI model catalog could mean for 5G

For you, the benefit would be indirect. If phones and networks lean on AI to manage connections, locate devices or save power, a shared catalog could stop them from re-sending the same models over and over, which the filing links to signal traffic, battery and computing costs. A phone could also ask for a model fitted to its situation, such as a certain area or a type of device.

For Samsung, the filing ties closely to work in 3GPP, the industry group that writes mobile network standards and has been studying how to handle AI models in 5G and 6G. A published application does not mean any standard will adopt it, though.

This is the 97th Samsung filing we've tracked since May in our 5G and network push watchlist, following one that blocks apps by location and one that predicts connection trouble.

Source: US 2026/0370563 A1 at the USPTO (PDF). Full patent text and figures are in the official publication.

Get our take in your Top Stories

Liked this breakdown? Add Patentlyze as a preferred source on Google, and our plain-English take shows up more often in your Top Stories the next time Samsung patent news breaks.

There are more where this came from

We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.

The drawings · 12 sheets

12 drawing sheets from US 2026/0370563 A1 · click any drawing to enlarge

Stages for building and running AI inside cellular networks, drawing sheet 1 from US 2026/0370563 A1
Sheet 1 Stages for building and running AI inside cellular networks
Successfully registering a new network AI model with the controller, drawing sheet 2 from US 2026/0370563 A1
Sheet 2 Successfully registering a new network AI model with the controller
What happens when the network rejects a new AI model, drawing sheet 3 from US 2026/0370563 A1
Sheet 3 What happens when the network rejects a new AI model
Submitting a new AI model without waiting for confirmation, drawing sheet 4 from US 2026/0370563 A1
Sheet 4 Submitting a new AI model without waiting for confirmation
Confirming an approved update to an active AI model, drawing sheet 5 from US 2026/0370563 A1
Sheet 5 Confirming an approved update to an active AI model
Handling a rejected request to update an existing AI model, drawing sheet 6 from US 2026/0370563 A1
Sheet 6 Handling a rejected request to update an existing AI model
Pushing an AI model update directly without an acknowledgment, drawing sheet 7 from US 2026/0370563 A1
Sheet 7 Pushing an AI model update directly without an acknowledgment
Alerting connected devices that an AI model has changed, drawing sheet 8 from US 2026/0370563 A1
Sheet 8 Alerting connected devices that an AI model has changed
Show the other 4 sheets
Notifying the server that an AI tool has changed, drawing sheet 9 from US 2026/0370563 A1
Sheet 9 Notifying the server that an AI tool has changed
Picking an approved AI tool from the network directory, drawing sheet 10 from US 2026/0370563 A1
Sheet 10 Picking an approved AI tool from the network directory
Rejecting a request when requested AI tools are unsupported, drawing sheet 11 from US 2026/0370563 A1
Sheet 11 Rejecting a request when requested AI tools are unsupported
Hardware parts that run the network communication and AI tools, drawing sheet 12 from US 2026/0370563 A1
Sheet 12 Hardware parts that run the network communication and AI tools
Patent filing page

Watchlists this filing belongs to

All watchlists →
Reader comments

No comments yet

Start the discussion

Real name or a handle, either is fine. Comments are read by a person before they appear, so allow a little time. Keep it about the filing.