Get the newsletter
Apple · AI / ML · Filed Dec 12, 2025 · USPTO published Nov 19, 2026 · This week's release

Apple Files Patent for Shortcuts That Pair Each Task With a Chosen AI Model

Apple describes shortcuts where every task gets its own assigned AI model, picked with security in mind, before your personal information is packaged into a question for it.

A smartphone touchscreen responds to input from a user's finger and an electronic stylus. Drawing from patent filing US 2026/0368225 A1.
A smartphone touchscreen responds to input from a user's finger and an electronic stylus.All 26 drawings ↓
Publication number
US 2026/0368225 A1(US20260368225A1)
Applicant
Apple Inc.
Filed · Published
Dec 12, 2025 · Nov 19, 2026
Inventors
Ayaka NONAKA, Duraid ABDUL, Andrea GOTTARDO, Daniel S. HSU, Xinyu HUANG, Nahush R. KULKARNI
Status when we published
Examiner raised objections; the company can respond
Document
10 claims
US classification
707/706
Examiner
AGHARAHIMI, FARHAD (Art Unit 2161)
Parent application
Claims priority from a provisional application 63808553 (filed May 19, 2025)

What Apple's AI-linked task shortcuts actually do

Today, a task shortcut is a saved job your digital assistant runs on request. It takes what you asked for, turns it into steps, and has your device carry them out. Apple's filing adds a twist: each task in a shortcut can be tied to a specific AI model, and running the shortcut means asking that model a question.

Suppose a shortcut needs your address book or a to-do list to answer something. The device first gathers that private data, arranges it into a neatly organized question, and sends it to the assigned model. The answer comes back and shows up as the shortcut's result.

The filing's stated reason for pairing tasks and models is security. Instead of one model handling everything, a task can be linked to the most secure model that is still capable of answering that kind of question.

From the filing · CLAIM 1
receiving an input including a request for a performance of a task shortcut; obtaining private user data required to perform a task of the task shortcut, wherein the task is associated with a machine-learned model; and performing the task of the task shortcut …

Translation: The system takes a request, gathers your private data, and runs it through a specific artificial intelligence model.

How a shortcut turns your data into a question

The filing covers two stages. In setup, you give a series of inputs that start a process for creating a task shortcut. That process attaches a task to the shortcut, then attaches one machine-learned model (an AI system trained on examples) from a set of available models to that task. Running the task then queries that model. The filing shows screens for managing these shortcuts.

At run time, the claims describe this sequence:

  • You ask for the shortcut to run.
  • The device gathers the private user data the task needs.
  • It builds a structured query (the data arranged into a consistent format the model can read).
  • It sends the query to the assigned model and receives a response.
  • It gives you an output based on that response.

The dependent claims add detail. Some of the private data can sit on the device itself while another part sits on a second device you own. The data can be all of one type (homogenous) or a mix of types (heterogenous). The structuring step can depend on the type of model, the type of task, or the types of data involved, so the same information may be packaged differently for different models.

From the filing · THE ABSTRACT
… associating a task with the task shortcut and associating a machine-learned model of a set of machine-learned models with the task, wherein performing the task causes the computer system to query the associated machine-learned model.

Translation: You can link a specific shortcut task to a chosen artificial intelligence model of your liking.

Why Apple wants the safest AI model per task

For you, the payoff is about trust. The filing's own pitch is that a task can use the most secure model able to answer its question, rather than one model for everything. It also says this makes the device more convenient, secure and intuitive, and saves battery because you get things done faster.

Keep in mind this is an application, and the examiner's first review rejected the claims as filed. That is a routine step that usually leads to amendments, so the final wording may change. The filing names iPhone, Apple Watch, iPad and other Apple devices as places its assistant could run, and the claims cover a computer system, a method and a storage medium.

This is the seventh Apple filing we've tracked since October in our AI models working together watchlist, adding to keeping follow-ups intact and routing questions to one model.

Source: US 2026/0368225 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 Apple 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 · 26 sheets

26 drawing sheets from US 2026/0368225 A1 · click any drawing to enlarge

Connecting your phone to cloud servers and web services, drawing sheet 1 from US 2026/0368225 A1
Sheet 1 Connecting your phone to cloud servers and web services
Internal parts and built-in apps that run the smartphone, drawing sheet 2 from US 2026/0368225 A1
Sheet 2 Internal parts and built-in apps that run the smartphone
How apps recognize and sort taps, swipes, and screen touches, drawing sheet 3 from US 2026/0368225 A1
Sheet 3 How apps recognize and sort taps, swipes, and screen touches
Exterior sensors, screen controls, and ports on the handset, drawing sheet 4 from US 2026/0368225 A1
Sheet 4 Exterior sensors, screen controls, and ports on the handset
Hardware and software layout for a desktop or laptop computer, drawing sheet 5 from US 2026/0368225 A1
Sheet 5 Hardware and software layout for a desktop or laptop computer
Basic steps apps follow to request or share data, drawing sheet 6 from US 2026/0368225 A1
Sheet 6 Basic steps apps follow to request or share data
Software parts that let apps speak to the underlying system, drawing sheet 7 from US 2026/0368225 A1
Sheet 7 Software parts that let apps speak to the underlying system
Sending requests and returning answers between programs, drawing sheet 8 from US 2026/0368225 A1
Sheet 8 Sending requests and returning answers between programs
Show the other 18 sheets
App icons arranged across a phone home screen, drawing sheet 9 from US 2026/0368225 A1
Sheet 9 App icons arranged across a phone home screen
Multi-finger gestures on a separate trackpad and screen, drawing sheet 10 from US 2026/0368225 A1
Sheet 10 Multi-finger gestures on a separate trackpad and screen
Front view of a smartwatch with physical buttons, drawing sheet 11 from US 2026/0368225 A1
Sheet 11 Front view of a smartwatch with physical buttons
Internal sensors and chips inside the smartwatch, drawing sheet 12 from US 2026/0368225 A1
Sheet 12 Internal sensors and chips inside the smartwatch
Server components that power voice assistant processing, drawing sheet 13 from US 2026/0368225 A1
Sheet 13 Server components that power voice assistant processing
How speech converts into structured actions and replies, drawing sheet 14 from US 2026/0368225 A1
Sheet 14 How speech converts into structured actions and replies
Organizing concepts to connect reservations with reminders, drawing sheet 15 from US 2026/0368225 A1
Sheet 15 Organizing concepts to connect reservations with reminders
The AI model turning your input into text responses, drawing sheet 16 from US 2026/0368225 A1
Sheet 16 The AI model turning your input into text responses
Building shortcuts with on-device AI actions and app menus, drawing sheet 17 from US 2026/0368225 A1
Sheet 17 Building shortcuts with on-device AI actions and app menus
Selecting AI models and scheduling automated notification alerts, drawing sheet 18 from US 2026/0368225 A1
Sheet 18 Selecting AI models and scheduling automated notification alerts
Choosing between cloud or local processing for shortcut steps, drawing sheet 19 from US 2026/0368225 A1
Sheet 19 Choosing between cloud or local processing for shortcut steps
Setting the exact output format for an AI query, drawing sheet 20 from US 2026/0368225 A1
Sheet 20 Setting the exact output format for an AI query
Running multi-step recipe workflows that generate ingredient lists, drawing sheet 21 from US 2026/0368225 A1
Sheet 21 Running multi-step recipe workflows that generate ingredient lists
Reviewing generated results and pairing screen time with cloud AI, drawing sheet 22 from US 2026/0368225 A1
Sheet 22 Reviewing generated results and pairing screen time with cloud AI
Filtering sports schedules and adding photo search queries, drawing sheet 23 from US 2026/0368225 A1
Sheet 23 Filtering sports schedules and adding photo search queries
Using cloud AI to scan travel photos for specific people, drawing sheet 24 from US 2026/0368225 A1
Sheet 24 Using cloud AI to scan travel photos for specific people
Linking a specific AI model to a custom shortcut task, drawing sheet 25 from US 2026/0368225 A1
Sheet 25 Linking a specific AI model to a custom shortcut task
Running a shortcut task by querying its paired AI model, drawing sheet 26 from US 2026/0368225 A1
Sheet 26 Running a shortcut task by querying its paired AI model
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.