Microsoft · AI / ML · Filed Feb 27, 2024 · USPTO published Sep 24, 2026 · Sep 24 release

Microsoft Patents an AI System That Predicts What You'd Type Into Work Forms

Filling out the same fields on the same forms every day is one of the quietest drains on office productivity. Microsoft is patenting a way for AI to do most of that typing for you, but with a catch: it only uses data you're actually allowed to see.

By Patentlyze Team · Updated Sep 24, 2026 · 3 min read

Data aspects related to predicting what a user would type into a form, including context, permissions, and content types. Drawing from patent filing US 2026/0289096 A1. — Data aspects related to predicting what a user would type into a form, including context, permissions, and content types.All 6 drawings ↓

Publication number: US 2026/0289096 A1(US20260289096A1)
Applicant: Microsoft Technology Licensing, LLC
Filed · Published: Feb 27, 2024 · Sep 24, 2026
Inventors: Ali ROUDAKI, Esin SAKA, Chuang HE, Miti JOSHI, Samarth GUPTA, Shankarr KALYANARAAMAN, Tanner DOZARK
Status when we published: Examiner raised objections; the company can respond
Document: 20 claims
US classification: 726/1
Examiner: GRACIA, GARY S (Art Unit 2499)
Parent application: Claims priority from a provisional application 63625552 (filed Jan 26, 2024)

How Microsoft's form-filling AI guesses your answers

Ever had to type your department, your project code, or your manager's name into yet another form, for the fifth time that week? Microsoft's patent describes a system designed to make that stop happening.

The idea is straightforward: instead of leaving the cursor blinking at an empty field, the system gathers context about you (your role, your recent activity, your organizational data) and sends it to an AI model. The AI reads that context and suggests what you'd probably type. You can accept the suggestion or ignore it.

The part that makes this more than a simple autocomplete is the guardrail layer. Before any data goes to the AI, the system checks whether you are currently authorized to access it. If your permissions have changed since the data was collected, it gets filtered out. The AI's response goes through a second check on the way back, making sure the suggested answer fits your job role and passes any rules your organization has set.

From the filing · CLAIM 1

gathering context data of the user; preprocessing some of the context data, the gathering or the preprocessing or both comprising verifying for at least a portion of the context data that the user has current permission to access said portion …

Translation: It checks what data you are allowed to see before using it.

How the system checks permissions before it predicts

Context gathering: The system pulls together relevant information about the user, such as their job title, department, recent documents, or calendar entries. This context is assembled to give the AI enough background to make a useful prediction.

Permission verification (preprocessing): Before the context data is packaged into a prompt and sent to the AI, the system verifies that the user has current authorization to access each piece of data. "Current" is the operative word: a permission that existed when data was originally collected may have been revoked since. Any data the user is no longer cleared for is stripped out before the AI ever sees it.

AI prediction: The cleaned context is wrapped into a prompt and submitted to an AI model (the patent covers both statistical models and large language model-style AI). The model returns a predicted value for the form field in question.

Postprocessing and validation: The AI's raw response doesn't go straight to the user. It passes through a second layer that:

Checks whether the suggestion respects responsible-prediction rules (filtering out answers that could be inappropriate or harmful)
Validates the answer against the user's role (so a junior employee doesn't get a suggestion that implies executive-level access)
Runs the answer against any organizational rules set by IT or compliance teams

Only after clearing all those checks does the system surface the suggestion to the user.

From the filing · THE ABSTRACT

Some embodiments postprocess the predictor response to enforce responsible prediction criteria, to validate against a user role, to validate against a rule, or a combination thereof.

Translation: The system filters its AI suggestions to make sure they follow workplace rules and security roles.

What this means for people who fill out forms all day

If you spend part of your day filling out internal forms, procurement requests, HR records, or project tracking tickets, this is the kind of change you'd notice immediately. The time saved on any single form is small. Across dozens of forms a week, the math adds up, and the frustration of repetitive data entry disappears almost entirely.

The permission layer is the part that actually makes this viable in corporate settings. Without it, an AI autocomplete could suggest information you shouldn't be seeing, or that's already out of date given a recent role change. Microsoft's run of enterprise-AI filings points toward a broader push to wire AI into the daily workflow tools millions of office workers already use, and a form-filling assistant that passes compliance checks is far more likely to get approved by IT than one that doesn't.

Microsoft's 15th filing we've tracked since June on AI assistants that remember you builds on one that stops chats from forgetting and one on auto-organizing folders.

Editorial take

The core idea here is straightforward and the value to an everyday user is real. If this works as described, you'd notice it the first morning you use it: less time hunting for the right cost center code, fewer fat-finger errors on fields you fill identically every week.

What separates this from a basic autocomplete is the two-checkpoint design. Permissions are verified before the AI sees your data, and the AI's answer is validated before you see it. That double-check structure is exactly what makes AI assistants viable inside organizations with compliance requirements, where a rogue suggestion can cause real problems.

The practical question is how well the system handles edge cases: a user whose role just changed, or a form with fields that have no clean historical pattern to draw from. The patent covers the architecture, but the daily experience will depend entirely on how the AI handles the messy reality of actual organizations.

Source: US 2026/0289096 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 Microsoft patent news breaks.

Add Patentlyze on Google One click.

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.