Big Tech’s Thirteen Most Interesting New Patents This Week (9/25/2026)

A weekly read of what the world’s biggest technology companies just told the U.S. patent office.

A full week by the numbers: we wrote up 286 filings, with Samsung out front at 48, then Google at 46, Qualcomm at 33, Apple at 32, and Sony at 25. OpenAI published exactly one. The full run lives under this week’s Big Tech patents; these thirteen stood out.

1. Disney Patents a Display System That Floats Real 3D Images in Open Air for Crowds

Every floating-image trick on the market works for one person standing in one spot. Disney has filed for a ring of optical stations whose light overlaps in midair to form a single real image a whole crowd can walk around.

Claim 1 is a set of display stations. Each has a light source, a half-mirror, and a retroreflector, with the half-mirror tilted at an offset angle. Light bounces off the mirror to the retroreflector, comes straight back, and passes through into the viewing space.

The stations are rotated relative to each other so their returned light overlaps. That overlap is the whole invention: one station gives a narrow cone, a ring of them gives a room. A second version folds the path with two more mirrors to shrink the footprint.

A real image is not a hologram illusion. The light rays physically converge at a point in the air, so it sits at a fixed spot in the room with no glasses and no phone. For a theme park queue, that is the difference between a demo and an attraction.

Three inventors, including Lanny Smoot, the Imagineer behind the HoloTile floor. Filed in May as a continuation of a June 2023 application, twenty claims. The open questions are brightness and how much floor space a ring of stations eats.

2. Google Patents a Search Tool That Listens to Background Audio Before You Type

The blank search box is the last moment Google does not know what you want. Google has filed for search suggestions built from whatever your phone’s microphone hears playing the instant you open the search bar.

Claim 1 starts when a search session opens with no characters typed. The phone captures background audio, the system identifies the media playing and pulls out entities from it, such as a show, a character, or an artist, then generates suggested queries and shows them before you type a letter.

This is not autocomplete. Autocomplete reacts to keystrokes; this reacts to the room. The pieces already exist on a Pixel, which recognizes songs playing nearby today, so the missing link is software that hands that recognition to search.

The product question is trust, not engineering. A search bar that listens before you ask will unsettle people unless the capture window, the retention, and the switch to turn it off are all obvious. The claims describe none of that.

Two inventors, filed in May as a continuation of a March 2024 application, twenty claims. The speed target is the hard part: the suggestion has to land before your thumb reaches the keyboard.

3. IBM’s New Patent Covers Keeping Human-Shaped Work Robots Charged Through the Floor

A humanoid robot that needs a person to plug it in is not autonomous. IBM has filed for a charging spot in the floor that locks a robot’s boots in place with electromagnets and pushes power up through the soles.

Claim 1 is a method with three alignments at once. A power receiver in the boots meets a transmitter in the floor, an RFID reader in the boots meets a tag in the floor, and an electromagnet in each boot meets one in the floor, locking the feet when switched on.

The RFID handshake is the clever piece. A passive chip costing pennies confirms the geometry is right before any power flows, and the magnets stop a forklift bump from sliding the robot off the pad mid-charge.

Nothing here is about the robot’s brain. It is the unglamorous logistics of keeping a walking machine running through a shift, which is exactly the layer that decides whether humanoids leave the lab.

Four inventors, filed in March 2025 as an original application, twenty claims. The claim is tied to boots on a floor, so a wheeled robot with a belly plate sits outside it.

4. Google Patents a TV That Recognizes Your Face and Loads Your Content Automatically

Switching profiles is the small tax every shared television collects. Google has filed for a display with a camera that recognizes whoever sat down and loads their content without a remote.

Claim 1 is short. A camera sends an image of a person in the room, a facial recognition module identifies the face, the system retrieves the user identifier tied to it, pulls that person’s content, and displays it automatically.

The filing also covers the reverse: when the camera sees you leave, the content stops. Walk in, your watchlist; walk out, the home screen or the next person’s profile. Presence becomes the login.

The obvious cost is a camera reading faces in the living room continuously, not once. The claims say nothing about consent or how face data is stored, and Google keeps filing on ambient presence, so this is not a one-off.

Two inventors, filed in May as a continuation of a June 2024 application, twenty claims. Google TV is the obvious home for it, if households accept the camera.

5. Sony Patents a Way to Turn Idle Game Controllers Into Extra Speakers

The second controller on the couch has a speaker and rumble motors and does nothing all night. Sony has filed for a console that spots which connected controllers are idle and routes audio to them.

Claim 1 is three steps. Detect every powered-on peripheral that has a vibration component, work out which ones are idle, and send audio signals to those. The audio can come out as sound, as rumble, or both.

Every DualSense sold since 2020 already has the hardware. The patent is a routing decision, so the path to a product is a firmware update and a settings toggle, not a new device.

The catch is fidelity. A controller speaker is tiny, so this probably works better as positional rumble cues scattered around the room than as real surround sound. As a free spatial layer for living-room play, that is still something.

Four inventors, filed in March as an original application, twenty claims. The filing does not limit the audio to game sound, so any stream the console handles could be redirected.

6. Tesla Patents a Rear Wing That Hides Until the Car Needs It

A big rear wing buys grip at speed and looks silly in a parking lot. Tesla has filed for a two-piece wing that sits flush with the bodywork until the car decides it wants downforce, then rises and tilts itself.

Claim 1 is a main wing plus a leading-edge slat. A deployment mechanism moves them between a stowed position, where the wing blends into the rear panel and the slat hides underneath, and a raised one. A rotation mechanism sets the angle of attack.

A controller runs both mechanisms in real time from vehicle parameters, trading drag for downforce as speed, steering, and braking change. The driver never touches a button and the car in the driveway looks clean.

The safety language is the tell. The filing describes how the wing behaves if it hits something, which is the paperwork a deployable exterior part needs to pass regulators. That reads like an engineer planning for production, not a concept sketch.

Two inventors, filed in August 2025 as a continuation of a March 2025 application, twenty claims. Tesla has several active-aero filings this year; a future Roadster is the obvious guess.

7. Samsung Patents an AI That Builds a Personalized Recap When You Resume a Paused Show

Five days after pausing a thriller, nobody remembers who betrayed whom. Samsung has filed for a system that tracks how you watched, then has a generative model write a recap of the exact scenes you saw.

Claim 1 logs consumption attributes during playback: stage changes in the story, key points about characters and plot, your rewinds and skips, your preferences, and whether the device suits a recap. When a pause passes a threshold, it ranks those signals by timing and by type before writing the summary.

The ranking is the specific part. Story key points and your own interactions outrank device status on both axes, so the prompt leans on what happened and what you did, not what screen you are holding. A short pause on a phone might get nothing; a three-day gap on a tablet gets the full recap.

It sits at the device layer, not inside one streaming app, so in principle it works across services. That is the part rivals would notice.

Five inventors, filed in March 2025 as an original application, twenty claims. The USPTO’s status record shows a notice of allowance in late August, so it is past the argument stage and headed for grant.

8. Disney Files Patent for an AI System That Watches Animals and Sends Alerts

A keeper cannot watch every enclosure, and reviewing a night of footage takes a day. Disney has filed for a camera system that runs a trained model over animal footage and pings staff when it sees behavior that matters.

Claim 1 is two steps and very wide. Process at least one image with a trained machine learning model to identify the animals or what they are doing, then transmit a notification to a device based on the result.

No species, no behavior list, no camera type is named. That is a choice: the same loop covers a live exhibit, a veterinary ward, or a research pen. The cameras already exist; the missing piece is a library of clips labeled by animal experts.

Animal Kingdom is the obvious home. A model trained on one species and a handful of behaviors, such as distress or feeding, is a small project with real value to the people responsible for those animals.

Three inventors, filed in March 2025 as an original application, twenty claims. The examiner has already pushed back once and Disney has answered, so a claim this broad is being argued right now.

9. Sony Patents a Physical Reaction Button Built Into Game Controllers

Reacting to a clutch save mid-match means opening a menu and killing the moment. Sony has filed for a controller button assigned to fire a reaction straight onto the screen while the game keeps running.

Claim 1 presents a game on a display, receives an input signal registering a reaction from a physical button assigned to that job, and shows a visual indication of the reaction on screen. The button also has at least one other function in the game.

That last clause is what makes it buildable. Controllers have a fixed number of inputs and players expect every one to do something, so the reaction lives on an existing button, perhaps a long press, instead of a new one.

Platforms compete on how well players can connect mid-game, and every menu between the moment and the reaction loses some of them. A like button for gameplay is a small idea with a clear reason to exist, which is more than most social features can say.

A single inventor, Yutaka Yokokawa, filed in May as a continuation of an October 2019 application, twenty claims. Seven years on the same idea suggests Sony wants this one to stick.

10. Waymo Patents an AI That Stops Its Robotaxis From Blocking Traffic

The signature robotaxi failure is a car parked in the worst possible spot with nobody inside to notice. Waymo has filed for a vehicle that asks a generative AI model whether it is in the way, and moves when the answer is yes.

Claim 1 is an autonomous vehicle whose data system builds a prompt for a generative model. The prompt holds spatial data, either raw sensor data or the arrangement of nearby objects, plus a request to judge whether the car is obstructing traffic. If the model says yes, the control system repositions the car.

A rule-based check can tell whether a spot is physically clear. A generative model can tell that a bus lane at rush hour is a different place from the same lane at midnight. Waymo is asking the car to reason like a dispatcher.

The hard part is latency. The filing does not say whether the model runs in the car or in the cloud, and on a city street that detail is the whole product.

Six inventors, filed in March 2025 as an original application, twenty-one claims. The USPTO’s own status record lists it as granted since mid-September, so this one is a patent in the full sense, not an application.

11. Samsung Patents an AI System That Tells Your Exoskeleton How Hard to Push You

Fitness trackers measure what you did. Samsung has filed for a powered wearable that adjusts how hard it pushes or resists your body, in real time, based on how close you are to the goal you set.

Claim 1 takes an exercise goal from the user, runs feature data from the workout through a neural network that outputs an estimated performance indicator, turns that into a torque parameter, and sends the parameter to the wearable so it generates that much force.

Powered exoskeletons exist in rehab clinics, running fixed programs. The difference here is a live loop: fall short of the target and the device assists more, beat it and the device backs off.

The AI half is a solvable problem with tools Samsung already owns. The mechanical half, a comfortable force-generating wearable, is years of hardware work, and the software has nowhere to run until it exists.

Three inventors, filed in May as a continuation of a December 2024 Korean international application, twenty claims. Samsung is staking out the territory early; nothing here suggests a gym-ready device is near.

12. Apple Patents Eye-Tracking That Scrolls Text When You Finish Reading

Reading in a headset still requires a deliberate scroll gesture. Apple has filed for a display that watches your eyes work through a paragraph and slides the next one into view the moment you finish.

Claim 1 shows two portions of text and tracks gaze. While the motion says you are reading the first portion, nothing moves. When the motion says you have finished, the text scrolls so the second portion lands exactly where your eyes are heading as they flick toward its beginning.

The trigger is the reversal, that left-to-right sweep followed by the jump back to the start of the next block. A rival that waits for a long pause instead would sit outside this claim, so it is narrow, but it is the right narrow thing for Vision Pro.

The filing also covers items that expand when you stare at them, head tilt plus gaze to switch screens, and an information panel that appears over whatever you are looking at. Gaze is being built into a full control vocabulary.

Eight inventors, including Bas Ording, who built the rubber-band scroll demo that sold Steve Jobs on the iPhone, and Steve Lemay, who now runs Apple’s human interface design group. Filed in May as a continuation of a March 2021 application, twenty-two claims.

13. Google Patents a Search System That Splits Your Query into Tasks and Learns From Your Clicks

Search for “how to start a vegetable garden” and the results cover soil, seeds, and beds in no useful order. Google has filed for a results page that splits the query into subtasks, ranks them, and reorders itself as you click.

Claim 1 sends the query through a multi-task model that finds a first and a second subtask, runs a separate search for each, and shows both on the page. When you interact with a second-subtask result, its score changes, and the page loads a third item for that subtask as you keep scrolling.

The page stops being a ranked list and becomes a feed that follows your behavior. Click nutrition links in a marathon search and the schedule articles you ignored stop arriving.

Two costs come with it. Results you saw ten seconds ago may not be there when you scroll back up, and clicks are noisy: a curiosity tap or a misleading headline becomes a vote the page acts on. Neither is addressed in the claims.

Three inventors, filed in June as a continuation of a December 2023 application, twenty claims. Subtask scores would also decide which sponsored results sit beside which content block, which is the part advertisers will read twice.

What ties them together

Six of the thirteen remove a step you used to take. Google listens before you type and recognizes your face before you pick a profile. Samsung decides whether you need a recap. Apple scrolls when it believes you are done. Google’s results page reorders itself around your clicks, and Waymo’s car judges its own parking.

The device is no longer waiting for an instruction. It is guessing, and the guess is the product.

The other column is hardware that stays out of sight until it is needed. Tesla’s wing lives inside the body panel. IBM’s charger lives in the floor. Sony’s spare controller becomes a speaker only when nobody is holding it, and its reaction button borrows an input that already exists.

Disney straddles both: a real image hanging in open air for a crowd, and an AI keeper watching the animals so a human does not have to.

Every guess in the first column is paid for with attention: a microphone at search time, a camera in the living room, a gaze tracker on your reading. None of these filings says how you agree to that. That argument is coming, and it will not be settled in a patent.

Patentlyze reads every new patent application that Apple, Google, Microsoft, Meta, Nvidia, Samsung, OpenAI, Qualcomm, Sony, Intel, AMD, IBM, Adobe, Salesforce, Tesla, Amazon, and Disney file at the USPTO, and writes a plain-English breakdown of the ones that say something about where products are heading. New filings drop every Thursday. The full archive lives in our patent library.