A new Microsoft AI patent application describes a system that can change what a person says during a live video, then redraw the speaker’s mouth so the edit cannot be seen. The filing, published by the United States Patent and Trademark Office on October 29, says the change would happen within seconds.
The application, titled “Near Real-Time AI Enhancement of Video or AR Presentations of Live Events”, was filed on April 29, 2025. Microsoft Technology Licensing is named as the applicant, with David Callaghan and Dinesh Kumar listed as the inventors. Microsoft has not said whether it plans to turn the idea into a product.
How the Microsoft AI patent system would work
An AI agent would listen to the live audio feed and pick out words to change, based on a set of rules and the viewer’s profile. Age, location and language could all shape the result. A generative AI system would then replace the words in both sound and picture, and smooth the audio and video around the swap so the join does not show.
The filing puts the delay at one to ten seconds, but that figure is a target written into the application, not a measured result. No demonstration of the system exists.
What the filing says it could do
The system could cut filler words such as “um”, remove slang, fix mumbled or mispronounced words, strip out offensive language, and remove background noise from a live feed.
A separate claim covers live translation. A viewer could hear a broadcast in their own language while the speaker’s lips appear to move in step with the translated words. The swaps could follow each viewer’s profile, and the filing also mentions swapping in facts drawn from a search.
Why the idea worries people
Because the edit would happen while the video streams, viewers would have no way of knowing what the speaker really said. That opens the door to misinformation and fraud, since a live broadcast could be altered before it reaches the audience.
The filing’s first claim says nothing about telling viewers an edit has happened. Only one later claim keeps the original recording, as a ground-truth record for review in a dispute. The system, in other words, is built to hide its own work.
Why it matters in Pakistan
Pakistan is already writing rules for AI-manipulated content. The Election Commission has proposed AI election rules for political parties, media organisations and election observers, after doctored clips spread during campaign season. Pakistan is also moving to teach schoolchildren about online safety, with a new cybersecurity council set to lead lessons in schools.
A tool that can rewrite live speech would test those rules in a new way, since it would alter video as it happens rather than after the event. Live video already plays a large part in how Pakistanis follow news, sport and politics.
What happens next
This is a patent application, not a product. No patent has been granted, and Microsoft has announced no service built on the idea. The patent office published the application on October 29, eighteen months after it was filed, which is when US applications are published as a matter of routine. The filing was flagged by patent tracker Patentlyze, which tracks its progress through examination.
The patent office has already pushed back: an examiner’s first review on July 29 rejected the claims as filed. That step is routine, and Microsoft can respond and amend the claims. The application is also limited to the United States, so protection elsewhere would need separate filings.
Microsoft has been pushing deeper into AI, from a major Copilot revamp to an AI system that diagnoses diseases. It also recently released Scope, a free tool to trace mistakes made by AI coding assistants, while OpenAI has teamed up with Hacktron to hunt for flaws in its AI systems. For now, the live-speech idea exists only on paper, but the debate it has started, about trust in live video, is already under way.
Frequently asked questions
What has Microsoft filed?
Microsoft filed US patent application US 2026/0341309 A1, for AI that rewrites a speaker’s words in live video and redraws their lips to match. It was filed on April 29, 2025 and published on October 29, 2026.
Has the patent been granted?
No. It is a published application, not a granted patent. The examiner’s first review on July 29 rejected the claims as filed, and Microsoft can respond and amend them.
How fast would the edit happen?
The filing gives one to ten seconds as an example, but that is a target in the application, not a measured result. No demonstration exists.
Why are people concerned about it?
Viewers would not know what the speaker actually said, since the edit happens as the video streams. That raises the risk of misinformation and fraud, and only one claim preserves the original recording.
