Microsoft has proposed an artificial intelligence system that could change what people say on live television and make their lips match words they never actually spoke.
Published on 29 October 2026, the US patent application describes technology that could alter speech and facial movements during live broadcasts. It raises a troubling question: could viewers still trust what they see and hear?
The proposed system would identify spoken words, replace selected phrases and generate matching video frames. Viewers could see a speaker apparently making statements that differ from their original remarks.
However, the technology remains a proposal described in a patent application. There is no publicly verified evidence that Microsoft has deployed the system on live television or demonstrated all its proposed capabilities.
How Microsoft's AI Could Rewrite Live Speech
The patent application, titled 'Near Real-Time Artificial Intelligence (AI) Enhancement of Video or Augmented Reality (AR) Presentations of Live Events', was filed by Microsoft Technology Licensing on 29 April 2025.
It describes AI agents that analyse audio and video recordings to identify words for removal or replacement. A generative AI system would then produce replacement speech and modify the accompanying footage. This includes generating video frames showing a speaker's lips moving in sync with the altered dialogue.
The system could also smooth transitions between original and modified footage, potentially making edits harder to notice. Proposed applications include television programmes, lectures, presentations, sporting events and augmented reality experiences.
Unlike conventional post-production editing, the technology could operate in near real time.
Microsoft describes a processing window of approximately one to ten seconds, comparable to delays used in some live broadcasts. However, this is a proposed operating window, not a verified performance result.
Why Microsoft's Patent Proposes Changing Spoken Words
Some proposed applications focus on improving speech without substantially changing its meaning. The patent describes removing filler words, correcting mispronunciations and modifying recordings to reduce audible or visible signs of speech impediments, including stuttering.
Live translation is another possibility. Someone speaking one language could appear to address viewers in another, with generated lip movements matching the translated words.
The system could also replace colloquial expressions, offensive language and words considered inappropriate for particular audiences.
However, the proposal goes beyond cleaning up speech. It describes obtaining replacement words through internet searches or databases containing information about the subject being discussed. Other examples include correcting errors or updating information to reflect newer facts.
These capabilities raise questions about editorial control. Who decides whether a statement is inaccurate, offensive or misleading? An automated correction might improve factual accuracy, but it could also change what the speaker intended to communicate.
Could Two Viewers Watch Different Versions of the Same Broadcast?
One of the patent's most striking features is its proposed ability to personalise what different viewers hear and see.
Microsoft describes user profiles containing information such as preferred languages, age, education and geographical background. These details could help determine which words are replaced and how the modified presentation appears.
Two people watching the same live event could therefore receive different versions of a speaker's remarks. One might hear the original statement, while another receives a translation or a version with certain expressions removed. Their videos could also show different lip movements.
Personalised subtitles and translated audio already allow audiences to experience content differently. Altering a speaker's visible mouth movements, however, introduces another concern. Without clear disclosure, viewers could mistake an AI-modified presentation for an accurate record of what was originally said.
Can Viewers Still Trust What They See on Live TV?
The implications could be particularly serious when someone's exact words matter. During a political interview, government announcement or breaking news report, replacing even a short phrase could change how audiences interpret a statement.
The patent includes a provision for preserving original recordings.
One claim describes storing unmodified audio and video as 'ground truth' for future reference. However, this does not guarantee that every implementation would retain the originals. The application also does not establish that viewers would automatically receive warnings about modifications.
Microsoft has previously explored related technology.
Its VASA-1 research project, published in April 2024, demonstrated AI-generated talking-face videos using a still image and an audio recording. That research was separate from the newly published patent and does not establish that Microsoft has implemented the proposed live-broadcast system.
The patent does not mean Microsoft is preparing to rewrite television broadcasts. It outlines potential technology rather than announcing an available product.
Nevertheless, it raises an important question for broadcasters: if AI can change someone's words and make their lips match, how will viewers know whether they are watching the original statement or an altered version?
