A weekly read of what the biggest tech companies, carmakers, toolmakers and aerospace firms just told the U.S. patent office.
We wrote up 406 filings this week, up from 213 last week. The jump comes from the 65 companies we started covering with this release, which brought 130 filings, led by Toyota with 16 and Hyundai & Kia with 15.
Among the original 17, Apple led with 76, then Google at 48, Microsoft at 35, Samsung at 33 and Nvidia at 20.
The full run lives on this week’s release page. These fifteen stood out, and five picks from the newly added companies close the post.
1. Google’s Health Tech Arm Patents a Swallowable Pill With a Needle That Dissolves After the Shot
Some drugs only work as shots because the gut breaks them down or barely absorbs them. Verily, Google’s health research arm, has filed for a capsule you swallow that settles against your stomach wall and injects the dose through a tiny needle that later dissolves.
Claim 1 is a capsule holding liquid medicine behind a seal. A trigger made of a material that dissolves in body fluids holds a plunger in place. Once the trigger dissolves, the plunger drives a hollow needle out of the capsule, and the medicine flows through a port in the needle.
The description puts the dose at 25 to 200 microliters and the trigger at roughly 5 to 10 minutes. One spring fires the needle 2 to 5 millimeters into the tissue, and a second spring pushes the dose. A heavy end and a floating end help the capsule settle upright.
The filing aims for the stomach because its wall averages about 5 millimeters thick, against about 1 millimeter for the intestine, which leaves a bigger safety margin. It says the dose could arrive 5 to 20 minutes after swallowing. How reliably the needle lands in a moving stomach is the open question.
One inventor, Joseph Dennis. It entered the U.S. in June 2025 from an international application filed in December 2023. Twenty claims, still waiting for examination.
2. A New OpenAI Patent Describes Earbuds That Pick Up Words You Only Mouth
Talking to an assistant on a quiet train means everyone nearby hears you. OpenAI has filed for earbuds that read words you mouth without a sound by feeling how your ear canal moves.
Claim 1 is a body that sits partly in the ear canal, a sensing system that detects canal motion caused by your mouth shaping words without voicing them, and a processor that turns that motion into a representation of the words.
Your jaw joint sits just in front of the ear, and the canal changes shape as it moves. The filing senses that with an accelerometer or with a strain gauge built into a soft silicone tip. A model guesses which of roughly 44 English speech sounds you made, then groups them into words.
The words could become a text, a command or speech played in another person’s earbud, and the filing pictures two people messaging silently. Training on your own mouthed sample words comes first. A wrong word in a silent text is the risk, since no sound gives you a chance to catch it.
Two inventors: Douglas Stuart Brodie and Daniel Boothe. Filed in May 2026 with priority from a provisional application of May 2025. Twenty claims, still waiting for examination.
3. Apple Files Patent for a Wearable That Hears Words You Never Say Aloud
OpenAI was not alone in chasing silent speech this week. Apple has filed for a head-worn device that reads the muscle movements of words you mouth and answers you privately.
Claim 1 is a housing worn on the head, a sensor whose signals reflect the nerve and muscle activity of speaking, and a processor that works out the substance of a conversation event and produces audio, text or a command.
In the main example, a laser lights your cheek and a detector watches the grainy speckle pattern the skin scatters. It shifts when facial muscles move by tiny fractions of a millimeter, and a neural network turns the shifts into words, adapting to a new wearer within minutes.
Dependent claims add assistant manners: a private reminder of an earlier topic, replies matched to your cadence and volume, and a check on whether a pause means you are thinking or done. The sensor has to hover near your cheek, so battery drain and alignment are the costs.
Four inventors, including Yonatan Wexler and Asaf Kagan. Filed in July 2026 as a continuation of an international application from September 2025. Twenty claims, still waiting for examination.
4. Google Patents an AI Podcast That Rewrites Its Script to Slip in Ads
AI-voiced shows need a way to pay for themselves. Google has filed for an AI podcast that answers a listener’s question by rewriting its script around content from a provider, such as a product pitch.
Claim 1 has a language model write a script from source material. When you ask a question, the system sends it with a slice of that material to a content provider, gets an item back and has the model write an updated script with a transition into it.
The filing’s example is a show about new-build homes and a question about washers and dryers under $2,000. The provider can return a product, a script outline or a preselected voice. The rewrite happens on the text before it becomes speech, and the next part can be written while the earlier part plays.
Claim 1 never says advertisement or podcast, so as written it could reach any AI-voiced show that pulls in outside material. The filing also describes spotting what it calls commercial intent breaks. Two listeners on the same topic could hear different content, even in different voices.
Two inventors: Lakshmi Kumar Dabbiru and Raghavendra Rao Donamukkala. Filed in May 2025 with no parent application, twenty claims. The examiner rejected the claims in a first review this month.
5. Google Files Patent for a TV Channel of AI-Made Videos Built Around You
Streaming apps recommend shows other people made. Google has filed for a channel in a TV app filled with videos an AI generates for you, shown under your name.
Claim 1 has a server gather your viewing preferences, context from your device such as its location or language setting, and demographic data such as an age bracket. It writes a prompt for a large language model, stores the video that comes back and streams it to your device.
A timer or a request starts the process, and one claim lets the prompt be built with no input from you. A second model can pull in news, social trends and sports scores; the example is a cricket fan during a tournament. An optional mode builds an avatar from your photos and copies your voice.
An ad stitcher can insert ads into the videos. Every personalized video costs computing power to make, where a row of recommendations costs almost nothing, and people who share less data get generic results.
Two inventors: Aravind Reddy Ayuluri and Prameela Kavya Thummuru. Filed in May 2025 with no parent application, twenty claims. The examiner has already issued a final rejection, which Google can answer by amending the claims or appealing.
6. Google Patents a Home Device That Spots Falls Without Recognizing Faces
Plenty of people will not put a camera in a bedroom, even one that could spot a fall. Google has filed for a home device that reads gestures and watches for falls using depth data that cannot identify a face.
Claim 1 is broad: take depth information from a sensor built from many photodiodes, recognize a gesture made by a person in its view and act on it. A second idea in the filing tracks a person’s center of gravity with infrared light and sends a notification based on it.
Tiny lasers flood the room with infrared light, and a time-of-flight sensor measures the returning light at roughly 30 to 60 frames a second. Software boxes each person, follows them, reads waves and swipes, and builds a 3D skeleton to find the balance point. A sharp shift in that point triggers a fall alert.
The filing names Nest Hub-style displays and Pixel phones as hosts and extends the idea to exercise form, slouching and restless sleep. It gives no error rate, and false alarms or missed falls matter most when an older person is on the floor.
One inventor, Aditya Shailesh Ghadiali. It entered the U.S. in March 2026 from an international application filed in September 2023. Twenty claims, still waiting for examination.
7. A New Apple Patent Would Let a Headset Remember Where You Left Your Keys
Retracing your steps to find a water bottle is a small, daily waste of time. Apple has filed for a headset that logs chosen objects and activities as its camera sees them, so you can ask later where something is.
Claim 1 is short: a device with an image sensor detects a context at a particular time, stores an entry pairing it with that time, and answers a later question about the context using the stored time.
A context can be an object, an activity such as brushing your teeth, or a state such as a locked door. Only items on a list you set get logged; you add one by holding it up to the camera. To save battery, it checks about once a second or once a minute.
Answers come with a time and a photo, such as the front door locked at 8:31 PM, so you know how far to trust them. A health app could notice a missed daily dose. The trade is a camera logging your day, and it only knows what it saw while you wore it.
Three inventors: Elizabeth V. Petrov, Devin W. Chalmers and Ioana Negoita. Filed in July 2026 as a continuation of an application from March 2024. Twenty claims, still waiting for examination.
8. Sony Patents an AI Game Soundtrack That Reacts to Your Heart Rate
Game music often loops the same track no matter how tense you are. Sony has filed for an AI composer that keeps reshaping a game’s score as your heart rate and the action change.
Claim 1 is an apparatus that starts generating a score on request and streams it to the console as it goes. During play it receives keywords selected partly from the player’s physiological data and updates the music each time a keyword arrives.
Developers write the list of mood words, such as cave, hostile or friendly. Sensors in a headset or controller can read heart rate, sweating and blinking. Seed values make the same keywords sound different per game or place, and a jump scare can force an abrupt change or hand off to prerecorded audio.
Sending a few words is far lighter than streaming the whole game state, and the filing says it keeps the AI easier to train. The examples use a PlayStation 5, a VR headset and a DualSense controller. Whether players want music that reads their pulse is open.
Five inventors, including Daniel Israel Pereira Jimenez and Lazaros Michailidis. Filed in July 2026 as a continuation of an application from February 2024. Twenty claims, still waiting for examination.
9. Apple Files Patent for Earbuds That Translate an Opera Without Muting the Orchestra
At a show in a language you don’t speak, you choose between watching the stage and reading the subtitles. Apple has filed for earbuds that speak a translation of the performer while the orchestra and the rest of the room keep playing in your ears.
Claim 1 has the earbud play a computer-generated translation of a performer’s voice and, at the same time, a pass-through feed of outside sound that keeps the background sounds and leaves out the performer’s voice.
Source separation software splits what the microphones hear into the performer, the wearer and everything else. The translating can happen on the earbud or on a paired phone. An optional extra mixes back a blurred, quieter copy of the performer’s voice, turned down more at high frequencies, so the stage still feels present.
The description puts the delay at a second or two, or under a second. Lifting one voice out of an orchestra, a chorus and a hall’s echo fast enough is the hard part, and the filing does not address translating sung lyrics.
Five inventors, including Yang Lu and Ashok Masilamani. Filed in April 2026 as a continuation of an application filed earlier that month, which appears at number 11 below. Twenty claims, still waiting for examination.
10. Apple Files Patent for Earbuds That Translate a Play While You Watch It
Two of the same inventors filed a second take on translated performances. Apple has filed for earbuds that keep the music and room sound while a computer voice speaks the actor’s lines in your language.
Claim 1 is a device with microphones, speakers and a processor. During a performance it picks up audio holding speech in one language plus ambient sounds, gets a translation while the show goes on and plays the ambient sounds with a computer-generated voice speaking the translation.
The original lines can be removed, turned down or blurred in time. The translated voice can be placed so it seems to come from the actor’s spot on stage, and shaped to resemble the actor’s pitch or tone, which helps you tell characters apart without looking away.
A second independent claim covers beamforming, combining several microphones to listen in one direction, with a listening zone that widens or narrows based on where two people stand and how the device moves. The examples run from opera and musicals to train and airport announcements.
Two inventors: Ashok Masilamani and Yang Lu. Filed in March 2026 with priority from a provisional application of May 2025. Twenty claims, still waiting for examination.
11. Apple Files Patent for Earbuds That Quiet the Room While They Translate
During a translated conversation you are listening to the computer voice, and street noise can bury it. Apple has filed for earbuds that turn up noise cancellation only while the translated voice plays.
Claim 1 runs noise cancellation at one level, obtains a computer-generated translation, plays it and, while it plays, raises the cancellation to remove more of the ambient noise. Other claims drop back afterward and set a third level while you speak, so your own voice sounds natural.
The boost can depend on pitch: deep sounds pass while mid and high sounds are suppressed. The filing says that blurs the background and reduces the ear pressure that sudden full cancellation can cause. A paired phone can send a start time so the earbuds ramp up before the translation begins.
The example is a tourist at a busy market stall in Madrid. A translated voice gives you no way to ask it to repeat itself, so a missed phrase can derail a conversation. Dependent claims also strip the original speaker from the pass-through sound, which ties it to the two filings above.
Five inventors, the same group as number 9. Filed in April 2026 with priority from a provisional application of May 2025. Twenty-two claims, still waiting for examination.
12. Google Files Patent for Ads That Blend Into AI Search Answers
A chatbot-style search answer has no obvious slot for an ad. Google has filed for a system that rewrites an advertiser’s content to match the tone of an AI answer and places it inside the text.
Claim 1 pairs an AI search assistant with two models. The first boils a preliminary answer down to a short proxy phrase. The phrase goes to a content server, and the second model rewrites the item that comes back so it fits the full answer.
The filing says ad servers handle short phrases well and long text badly, so the proxy phrase runs fewer than six words. The lookup runs while the full answer is written, so you are not kept waiting. A check leaves links out when the answer has nothing to do with a product or service.
The example is a question about making vegetable juice at home, answered with a shopping link on each ingredient. The filing says people learn to ignore anything that looks like a typical ad. Everything here is server software, so a text-only version is the short route to a feature.
Four inventors, including Matthias Heiler and Dorothea Wiesmann Rothuizen. Filed in May 2025 with no parent application, twenty claims, still waiting for examination.
13. OpenAI Files Patent for Locking an AI Coder Offline Before It Touches Your Code
Letting an AI agent work on private code raises one question first: what can it reach? OpenAI has filed for a sealed workspace that installs everything a project needs, cuts the internet and only then lets the AI start coding.
Claim 1 is short. Start a container with an internal link to an AI model and a network link to the public internet, download the packages the repository needs to build, and then disable the network link.
While the internet is still on, the container copies the repository, using stored passwords if it is private, installs its packages and runs any startup script. The AI then works through a terminal and a file editor. At the end the system saves a summary of changes and a command log.
The description says a stray command could otherwise upload code to an outside server or fetch harmful software. It also describes several workspaces running at once, each building a different piece, with a person reviewing the changes. An alternative swaps the hard cutoff for a proxy that allows only certain websites.
Three inventors: Fotios Chantzis, Channing Conger and Joshua Ma. Filed in September 2025 with priority from a provisional application of May 2025. Twenty claims, still waiting for examination.
14. Apple Files Patent for a Phone Feature That Waits on Hold for You
Hold music keeps you tied to the phone in case a person finally picks up. Apple has filed for a phone that listens to the call for you and tells you when a live agent comes on.
Claim 1 has the device analyze a call’s audio to decide which of several call states it is in, such as on hold or talking with a person, and push a notification that changes what the screen shows.
After about 30 seconds of music or silence, the phone can offer to take over and mute the call. Transcribed speech is sorted word by word into recordings or a human agent. When a person speaks you get an alert, on a nearby watch if needed, and the agent hears that you are coming back.
Another claim predicts hold before the call starts, using a scrambled list of numbers that have put callers on hold before. The processing is described as running on the device. Accuracy is the risk: a phone that mistakes a recorded greeting for a person cries wolf.
Six inventors, including Mahak Goindani and Jae Woo Chang. Filed in January 2026 with priority from a provisional application of May 2025. Twenty claims, still waiting for examination.
15. Apple Patents a Call Screen That Hears Hold Music and Offers a Hold Button
The call screen shows the same buttons whether you are talking to a friend or listening to a jingle. Apple has filed for a call screen that shows a Hold button only when the audio sounds like hold music, a recording or silence.
Claim 1 has the system check the audio of an active call. If the call meets a set of hold criteria, a Hold option appears on the call screen. If it does not, the option stays hidden.
The criteria can include music or an automated message from the other side, nobody speaking, or you being silent or on mute. Tapping Hold stops the call audio and shows an on-hold indicator next to your lock screen, home screen or another app. A transcript can show what the other side says meanwhile.
The filing calls today’s call screens cumbersome, with taps that waste time and battery, and it names the iPhone as an example device. Guessing wrong about hold music, in either direction, is the main way it could annoy.
Two inventors: Jae Woo Chang and Edward Chao. Filed in January 2026 with priority from a provisional application of May 2025. Sixteen claims, still waiting for examination.
What ties them together
Two companies went after silent speech in the same release. OpenAI reads the ear canal from inside an earbud, and Apple reads the cheek with laser light from a head-worn device. Both picture two people messaging without a sound, and both have to get the words right with no voice to fall back on.
Apple placed seven of the fifteen, in two clusters: three filings on translation earbuds from overlapping inventors, one a continuation of another, and two on hold calls filed the same day. Google placed five, counting Verily, and two of them put ads inside AI output: the podcast script and the search answer.
Several ask for a lot of personal data. Google’s TV channel uses your viewing habits, location and age bracket, and Apple’s headset logs what its camera sees through the day. Google’s fall detector goes the other way, built on depth data that cannot identify a face.
None of the fifteen is a granted patent. Google’s podcast drew a first-round rejection and its TV channel a final rejection. Four are continuations, two entered the U.S. from international applications filed in 2023, and six claim priority from provisional applications filed in May 2025.
New this week: five from the companies we just added
Nov. 19 was our first release with 65 more companies across cars, gear, flight and space, and emerging tech. Here is what changed and who we added. Five of their first 130 filings:
GM Files Patent to Heat EV Cabins With Heat From Your Brakes
GM says cold weather can cut an EV’s range by 30 or 40 percent, largely because the electric heater draws on the drive battery. Its filing pipes air warmed by the brakes into the heating system and switches the electric heater off when it is cold outside. Filters catch brake dust on the way.
SpaceX Files Patent for Internet Dishes That Hop Satellites Before a Tree Blocks Them
SpaceX has filed for satellite internet dishes that switch to a backup satellite on their own. Each dish gets a schedule minutes ahead listing at least two satellites for every short time slot, each tied to a different ground station, and it can map nearby trees to move before one blocks the signal.
Figure AI Patents a Backpack Rig That Lets Workers Teach Humanoid Robots
Figure AI has filed for a backpack frame with two jointed arms that follows a worker’s arms and records every joint, so humanoid robots can learn from real jobs such as moving sheet metal. The same rig can steer a robot live, with a target delay under 20 milliseconds.
Lutron Patents a Clip-On Frame That Turns Old Wall Switches Into Smart Controls
Lutron has filed for an adapter frame that screws over an existing wall switch so a battery-powered smart panel can clip on with no new wiring. Reversible brackets handle two-switch wall plates where one switch may sit deeper than the other. Renters who cannot rewire are the obvious audience.
Motorola Patents a Phone That Only Hears Voices Its Camera Can See
Motorola has filed for a phone whose microphones only listen inside the area its camera can see, so voices behind or beside it get ignored. When two devices hear the same command, the filing uses signals such as where you are looking to pick which one answers.
Patentlyze reads every new patent application from 82 companies, from Apple, Google and Microsoft to Toyota, Ford, Milwaukee, Airbus and SpaceX, 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.
Get lists like this in your inbox
We read the new patents from Big Tech, carmakers and more and send you the ones worth knowing. Plain English, free, every week.

