Microsoft has filed a patent application for an artificial intelligence system that can change what a person says while they are still speaking on live video, and then redraw their mouth so the edit cannot be seen. The change would take only a few seconds, so viewers watching a live broadcast would have no way of knowing that the words they heard were not the words spoken.
The filing, published by the United States Patent and Trademark Office on 29 October 2026, is listed as US application 20260341309 and titled Near Real-Time AI Enhancement of Video or AR Presentations of Live Events. Microsoft Technology Licensing is the applicant, with David Callaghan and Dinesh Kumar listed as the inventors. It was filed on 29 April 2025 and became public eighteen months later, when American patent applications are routinely published. No patent has been granted, and Microsoft has not announced any product based on the idea.
What makes the filing stand out is that it is designed for a live stream, not a recording. An AI agent would listen to the audio as it arrives, pick out words to change, and replace them, while a second part of the system rebuilds the speaker’s mouth movements so the new words look natural. The filing gives a delay of one to ten seconds as an example, similar to the short delay broadcasters already use to remove strong language from live television. That figure is a target in the application, not a measured result, and the filing includes no demonstration of the technology at that speed.
What the System Could Change
The application lists live translation first among the possible uses. A viewer could hear a speech in their own language while the speaker on screen appears to be speaking it, with the lips moving to match. The same system could also remove filler words such as um and ah, take out stutters and mispronounced words, replace regional slang with more standard phrasing, and remove offensive language automatically.
The choices would not be the same for everyone watching. The words to be changed could be picked using a set of rules or a profile of the viewer, taking account of age, location and language. Two people watching the same live event could therefore hear two different versions of what was said. The system could run on phones, tablets, cameras and augmented reality glasses fitted with on-device chips.
Why It Raises Concerns
The same ability is also the reason the filing has drawn attention. If a system can put new words into a speaker’s mouth during a live broadcast and make the change invisible, a viewer would have no reason to doubt what they saw and heard. A public figure could appear to say something they never said, in footage that looks untouched. That risk is sharper for live video than for recordings, because people tend to trust a live broadcast more.
The filing does include one safeguard. A single dependent claim covers keeping the original recording, so the unedited audio and video would be preserved as a ground truth record that could be checked later if a dispute arose. But the published summaries of the application note that it contains no notice to the viewer, so there is no clear sign that the audience would be told an edit had taken place.
There is also the question of who decides what gets changed. The words could be chosen by preset rules or by a viewer profile, which puts a great deal of power in the hands of whoever sets those rules. A broadcaster, a platform or a government could in theory decide which words are acceptable and which are not, and the audience would never see the choice being made.
What Happens Next
For now, the system exists only on paper. A patent examiner rejected the claims as filed on 29 July 2026, a routine step at this stage, and Microsoft can respond by amending the claims and arguing for them. Even if a patent were eventually issued, it would cover only the United States, since the published record lists no related filings in other countries.
Microsoft has not said whether it plans to build the system, and there is no launch date, no public demonstration and no announcement of a product. The company has, however, been pushing further into artificial intelligence across its business. It recently carried out a major revamp of Copilot, its AI assistant, and it has developed an AI system for medical diagnosis that researchers say can diagnose diseases more accurately than doctors in some tests.
A patent application fits that wider pattern of investment, but a filing is not a promise. Large technology companies file thousands of patents every year, and many of them never become products. The filing will now move through the patent process, where Microsoft can amend its claims and make its case. If a system like this ever reaches a real broadcast, the audience will need to know when the words they hear are the speaker’s own and when they have been rewritten by a machine.
