Feature · AI twins
Jensen Huang’s AI Twin: What a Digital Founder Can and Cannot Do
Nvidia spent years building computer-generated versions of its founder, from 14 seconds of a digital Jensen Huang in a 2021 keynote to Toy Jensen and a lifelike avatar at Computex 2024. Then Huang resigned in a single day with no successor named. This is what an AI twin of a chief executive can keep for the people who depend on them, what it can never do, and who should control it.
The short answer: a Jensen Huang AI twin already exists in pieces, and Nvidia built most of them itself. In April 2021 a computer-generated Huang delivered 14 seconds of a keynote from a digital copy of his kitchen, in November 2021 a talking avatar called Toy Jensen answered engineers’ questions live, and at Computex 2024 an avatar of Huang narrated chapters of his keynote. In March 2026, Huang told Lex Fridman that his inbox and “everything I’ve said” had been “collected and becoming my AI.” On November 12, 2026, he resigned as Nvidia’s chief executive, effective immediately, and the board named an interim CEO and opened a search, though by November 15 the Financial Times was reporting that he had discussed a comeback (Reuters). In our view, none of those digital Jensens could run Nvidia. A twin built from a founder’s own reasoning, under his own rules, could still keep his judgment within reach of the people he trusts, and that is the question this feature answers for every CEO and founder.
Number | What it is | Source |
|---|---|---|
14 seconds | of Nvidia’s April 2021 keynote delivered by a computer-generated Huang | |
~100,000 | peak live viewers of a deepfake Huang stream in October 2025 | |
33 years | as Nvidia’s only CEO, ending on November 12, 2026 | |
75+ | presentations given by Reid Hoffman’s AI twin since 2024 | |
$129 million | allegedly stolen with deepfaked CEOs, in a November 2026 federal case | |
23% | of boards with a formal written CEO succession plan |
Nvidia built a digital Jensen years before he left
For an hour and 48 minutes in April 2021, Jensen Huang presented Nvidia’s GTC keynote from what looked like his own kitchen. For 14 of those seconds, from 1:02:41 to 1:02:55, the man on screen was a computer-generated clone standing in a computer-generated copy of the room. Nvidia revealed it four months later in a post on its blog: its teams had scanned Huang’s face and body to build a 3D model, trained an AI to copy his gestures and expressions, and rendered the result in Omniverse, its 3D simulation platform. Nvidia’s first wording led many outlets to report that the whole keynote had been fake, and VICE recorded that the company then updated its post to say only 14 seconds were animated.
Seven months later, at its November 2021 GTC, Nvidia launched Omniverse Avatar, a platform for interactive AI avatars, and showed Toy Jensen, a small animated version of its founder that could hold a conversation. Nvidia’s making-of account says its Riva software handled speech recognition and turned answers into speech, while its Megatron 530B language model understood the questions and wrote the replies. When an engineer stumbled over a scripted question, Toy Jensen answered off script that “the greatest are those who are kind to others,” and Nvidia used the line in the keynote. Toy Jensen was the first digital Jensen that talked back, and it is the one most people mean when they search for a Jensen Huang AI clone.
Nvidia then turned the idea into products. On June 2, 2024, it made its ACE software for digital humans generally available to companies including Dell and ServiceNow, and the same Computex keynote featured a digital Huang of its own. Synthesia says it built the avatar on its EXPRESS-1 model, that it narrated several chapters of the keynote over video demonstrations, and that it was “nearly indistinguishable from the real Jensen Huang.” That description is the vendor’s own, but the avatar was on Nvidia’s stage for anyone to judge.
As far as the public record shows, every digital Jensen was a performance. Each one was a face, a voice and a script, or a language model speaking in a borrowed style for a few minutes on stage. None was built to hold what Huang knew about running Nvidia or how he weighed a decision, and none had any role on the day he left. That difference, between a likeness of a person and a twin of their judgment, is the subject of the rest of this feature.
Every digital Jensen, in order
Five years of Nvidia copying its own founder, one copy it did not make, and the day the real one left.
April 2021
A clone in the kitchen
For 14 seconds of a keynote lasting an hour and 48 minutes, the Huang on screen is computer-generated. Nvidia reveals it in August.
Nov 9, 2021
Toy Jensen talks back
Nvidia launches Omniverse Avatar and shows Toy Jensen, an animated Huang that answers engineers’ questions live.
June 2, 2024
A digital Huang at Computex
Nvidia releases its digital-human software, and an avatar of Huang built by Synthesia narrates chapters of his keynote.
July 29, 2024
“Everybody will have an AI assistant”
Huang says so at SIGGRAPH, where Nvidia previews James, a digital human the public can talk to.
Jan 6, 2025
The HR department of AI agents
At CES, Huang calls AI agents a digital workforce, Fortune reported, and Nvidia shows a desktop avatar, Project R2X.
Oct 28, 2025
A deepfake Huang outdraws the real one
A fake “NVIDIA LIVE” stream pushes a crypto scam during the GTC keynote in Washington, peaking near 100,000 viewers, PCWorld reported.
Jan 27, 2026
No clear successor
Bloomberg reports that Nvidia lacks a clear successor for the chief executive who built it.
Feb 25, 2026
Nvidia warns about “transfer of knowledge”
Its annual report says the company is “highly dependent on the services of our longstanding executive team.”
Mar 23, 2026
“Becoming my AI”
On the Lex Fridman Podcast, Huang says he does not believe in succession planning and describes his digital record becoming “my AI.”
Nov 12, 2026
Huang resigns in a single day
Nvidia says he has resigned as CEO and director, effective immediately. Colette Kress becomes interim CEO, and the board starts a search.
Nov 13, 2026
Musk backs a petition to reinstate him
“A petition is underway to reinstate Jensen and I’m signing it too,” Elon Musk posts. Nvidia says nothing about reversing the resignation.
Nov 14, 2026
“A handshake with Jensen”
Musk posts that many of Nvidia’s supply deals were “a handshake with Jensen, not a written contract,” and that some “critical deals will be lost” without him.
Nov 15, 2026
Huang weighs a comeback
The Financial Times reports that Huang has discussed a return as CEO with confidants, Reuters reported. Only Nvidia’s board can bring him back.
The deepfake Jensen that outdrew the real one
On October 28, 2025, while Huang gave the keynote at Nvidia’s GTC conference in Washington, D.C., a YouTube channel calling itself “NVIDIA LIVE” streamed a deepfake of him. PCWorld reported that it peaked near 100,000 live viewers, more than eight times the audience of Nvidia’s real stream, and for part of the keynote ranked above the real stream in YouTube’s search results. Engadget reported that the fake Huang announced “a crypto mass adoption event that ties directly into NVIDIA’s mission” and told viewers to scan a QR code and send cryptocurrency that would supposedly come back as Bitcoin. Outlets give peak counts from about 95,000 to about 120,000, but every report we read agrees that a fake Huang briefly drew a bigger crowd than the real one.
In our view the fake worked for the same reason Nvidia’s own avatars impressed people. Huang’s keynotes, interviews and earnings calls are public, so anyone who wanted to copy his face, voice and manner had years of material. The two stories are mirror images. In 2021 Nvidia controlled its digital Huang, approved its script and chose when it appeared, while in 2025 a stranger controlled one and it said whatever the scam needed.
We think that is the central question for any chief executive deciding whether to build an AI twin: who decides what it says, who may talk to it, and what happens to it when the founder is no longer there to object. Those questions apply long before death, as the deepfake shows, and they become urgent after it. We look at the law further down, and at who owns an AI clone in a separate guide.
“Becoming my AI”: what Huang said about his own twin
The closest Huang has come to describing an AI twin of himself came on the Lex Fridman Podcast, in an episode published on March 23, 2026. Talking about the far future, he said he would one day put a humanoid robot of himself on a spaceship. Then he described the rest of himself: “Take all my inbox, take everything that I’ve done, everything I’ve said. You know, it’s been collected and becoming my AI.” When the time came, he said, that AI would be sent at the speed of light to catch up with the robot (transcript, 2:22:50 to 2:24:00).
He was speaking playfully, and nobody should read a product plan into it. The idea at its center is serious all the same. A lifetime of email, meetings and speeches already exists as a digital record, and an AI trained on that record can answer in a person’s voice and style, which is the plain definition of an AI twin.
The same conversation shows why Huang never built one for Nvidia’s succession. About five minutes earlier he said, “I’m famous in saying that I don’t believe in succession planning,” adding that this was not because he thought he was immortal. His alternative was to “pass on knowledge, information, insight, skills, experience as often and continuously as you can,” which is why, he said, “every single meeting is a reasoning meeting.” In January 2026 he was quoted saying that almost 60 of his direct reports could be world-class chief executives elsewhere and that “there will never be one like me,” Wccftech reported. In an interview with Bloomberg’s Emily Chang, recorded weeks before he left, he said he was “really hoping that the next CEO of the company’s already at the company,” BusinessToday reported.
Our reading is that Huang’s method was itself a kind of twin, spread across roughly 60 people. Every executive who sat through his reasoning carries part of it, which helps explain why Nvidia’s bench is considered deep. The weakness is that nobody holds all of it, and none of it sits in a form that the next leader or a new director can simply ask. A twin built from his reasoning would have kept the whole of it in one place, available to the people he chose, under rules he set.
Jensen Huang on AI twins, agents and his successor
Six things Huang said on the record between 2024 and 2026, each with where he said it.
Lex Fridman Podcast · Mar 2026
“It’s been collected and becoming my AI”
- His inbox
- Everything he said
Read the transcriptHuang described his digital record, from his inbox to everything he has said, as already turning into an AI of himself.
Lex Fridman Podcast · Mar 2026
“I don’t believe in succession planning”
- Not because he is immortal
Read the transcriptHis alternative was to pass on what he knows continuously, by reasoning out loud in every meeting.
Bloomberg interview · 2026
“The next CEO has to be hired”
- Reported by BusinessToday
Read the reportRecorded weeks before he resigned. He said he hoped the next chief executive was already at Nvidia.
SIGGRAPH · July 2024
“Everybody will have an AI assistant”
- With WIRED’s Lauren Goode
Read Nvidia’s accountHuang said every company and every job would have AI help, with digital agents working alongside people.
Bg2 podcast · Oct 2024
“100 million AI assistants”
- Reported by Cryptopolitan
Read the reportHe said he hoped Nvidia would one day be a 50,000-employee company with 100 million AI assistants across its groups.
CES · Jan 2025
“The HR department of AI agents”
- Reported by Fortune
Read the reportHe said every company’s IT department would onboard, maintain and improve a workforce of digital agents.
He left in one day, with no successor named
On November 12, 2026, Nvidia announced that Huang had resigned as president, chief executive officer and a director, effective immediately, and its SEC filing gave no reason. Colette Kress, chief financial officer since 2013, became interim CEO and keeps her finance job until a permanent successor is named. Harvey C. Jones, a director since 1993, became the board’s first chair and said Nvidia’s operations, customer commitments and product roadmap should “continue without interruption.” The board will consider internal and external candidates and set no timetable, and our companion report on who will be Nvidia’s next CEO follows the search.
Nvidia had warned its investors in plain words. Its annual report said the company is “highly dependent on the services of our longstanding executive team,” and that failing to ensure “effective succession planning, transfer of knowledge, and smooth transitions” could hurt its long-term success. A full-text search of SEC filings finds the “longstanding executive team” line in every Nvidia annual report from 2022 to 2026, and Bloomberg reported in January 2026 that Nvidia lacked a clear successor.
The phrase in that warning that matters most here is “transfer of knowledge.” A successor can be hired in months, and boards do it all the time. The judgment of a founder who ran a company for 33 years is far harder to hand over, because most of it was never written down. When Huang left in a day, the parts of his thinking that lived only in his own head left with him.
Outside Nvidia, the reaction showed how much one person had come to stand for the company. Elon Musk replied “That would be an honor” on November 11 to a post urging him to hire Huang, CNBC reported, and on November 13 he posted that “a petition is underway to reinstate Jensen and I’m signing it too.” The petition turned out to be the Stand with Jensen open letter, which Musk did not name at first, Tech Startups reported. The next day he made the point that matters most for this page: many of Nvidia’s supply deals, he wrote, were “a handshake with Jensen, not a written contract,” and without Huang “at least some of those critical deals will be lost.” On November 15, the Financial Times reported that Huang has discussed a return with confidants, Reuters reported, a possibility we cover in Can Jensen Huang come back?, and Nvidia has said nothing about reversing the resignation. Whether or not he returns, the week showed how much of a company’s knowledge, and even its promises, can sit with one person and nowhere else.
What an AI twin of a CEO can and cannot do
This is our assessment, based on how executive twins are used today. A twin is a tool that its owner controls, and it carries no legal authority of its own.
- Can
Answer in the founder’s own words
Trained on what a founder has said and written, a twin can explain how the founder approached a pricing call, a hire or a bet on a product, using the reasoning the founder actually gave.
- Can
Speak in the voice people know
Executives already use voice clones on earnings calls and in client videos. A twin that answers in the founder’s own voice carries weight with a team that has just lost its leader.
- Can
Stay available, under rules
With the right controls, a twin can stay available to the people the founder chose after he retires, leaves or dies, and only to them.
- Cannot
Run the company
A twin cannot hold office, sign contracts, vote shares or owe duties to shareholders. Those belong to people, and in a public company the board picks them.
- Cannot
Make a decision the founder never made
A twin can report how its owner reasoned in the past. When the facts change, as they did at Nvidia on November 12, the decision belongs to the people in charge now.
- Cannot
Be trusted without its owner’s consent
A copy built without permission is a deepfake, however good it looks. The fake Huang of October 2025 showed that a familiar face and voice prove nothing about who is speaking.
Can an AI replace a CEO?
Some chief executives think a large share of their own job could go. In a 2023 poll by edX and Workplace Intelligence, 49% of CEOs said most or all of their role should be completely automated, HC Magazine reported. Reid Hoffman, the co-founder of LinkedIn, was asked in 2026 whether his AI twin could replace him and answered that it could, according to Wall Street Journal reporting summarized by Expansión. He predicted that within 10 years every company with more than 50 employees would have virtual versions of its executives.
The people who would work alongside those twins are less sure. Researchers at Carnegie Mellon and Emory interviewed 23 managers and workers for a paper titled “When Your Boss Is an AI Bot” and found “a fundamental tension between the promise and risks of manager clone agents,” The Register reported. IMD professor Michael Watkins asked in September 2026 whether leadership can be delegated to a digital self, citing Klarna’s and Meta’s AI chief executives.
Our answer to whether an AI can replace a CEO is no, and the Nvidia case shows why. A chief executive is accountable to a board, to shareholders and to the law, and only a person can carry that accountability. On November 12, Nvidia’s board needed to know who would be in charge the next morning, and only a human answer would do.
What a twin can replace is the large part of a chief executive’s week spent repeating what he already thinks. That includes answering the same questions from new staff, explaining decisions made years ago, briefing a director who joined last quarter, and presenting results that follow an agreed script. Huang himself predicted that IT departments would become “the HR department of AI agents.” If agents become the workforce he described, we think a founder’s own twin is the agent that matters most, because it is the only one that knows how the founder thinks.
The CEOs and founders who already use AI twins
Every case here is on the public record, with who reported it. Most are avatars that present a script, and a few are twins that answer questions.
Zoom · May 2025
Eric Yuan
- Earnings call avatar
- Zoom Clips
An AI avatar delivered Yuan’s opening remarks on Zoom’s quarterly call. “I am proud to be among the first CEOs to use an avatar in an earnings call,” it said, TechCrunch reported.
Klarna · Dec 2024 and May 2025
Sebastian Siemiatkowski
- Results videos
- Introduced itself as AI
An AI avatar of Klarna’s chief executive presented its first-quarter 2025 highlights in a short video, after a first version did the same in December 2024, Entrepreneur reported.
LinkedIn co-founder · since 2024
Reid Hoffman
- Reid AI
- 20 years of his words
His twin is a custom GPT trained on about 20 years of his books, speeches, podcasts and interviews, Popular Science reported. It has given more than 75 presentations and speaks 74 languages, according to WSJ reporting carried by Expansión.
Meta · reported April 2026
Mark Zuckerberg
- Photorealistic
- Built for staff
Meta is building a photorealistic AI Zuckerberg, trained on his mannerisms, tone and public statements, so that staff “feel more connected to the founder,” the Financial Times reported, as carried by The Irish Times.
Customers Bank · April 2026
Sam Sidhu
- Earnings call
- Nearly half an hour
An AI clone built with OpenAI read Sidhu’s prepared remarks on the bank’s quarterly call for nearly half an hour before he revealed it to analysts, Fortune reported.
Uber · reported March 2026
Dara Khosrowshahi
- Dara AI
- Pitch rehearsal
Some Uber teams built a “Dara AI” and rehearse their presentations on it before they meet him, he told The Diary of a CEO podcast, HC Magazine reported.
JD.com · April 2024
Richard Liu
- Livestream host
- 20 million views
An AI avatar of JD.com’s founder hosted a sales livestream that drew more than 20 million views in its first hour, the company said, at a tenth of the cost of a human host, Yicai reported.
Greif and Quantum Health · 2025 to 2026
BalaBot and Virtual Brian
- HR chief
- Chief executive
Greif’s HR chief Bala Sathyanarayanan has a twin called BalaBot that has interacted with about 3,300 employees, and Quantum Health’s chief executive Brian Hartzer has “Virtual Brian,” according to WSJ reporting carried by DiarioBitcoin.
UBS · since January 2025
Research analysts
- More than 36 of about 720
UBS uses AI avatars of its analysts for client videos, after studio time had limited it to about 1,000 videos a year, Fortune reported.
What CEO twins are used for today
Read those cases together and the uses fall into three groups. The first is presenting: Yuan, Siemiatkowski and Sidhu each let an avatar deliver words they had approved, and part of the point was the demonstration itself. The second is access. Meta’s Zuckerberg twin, Uber’s “Dara AI” and Greif’s BalaBot let employees reach a leader’s thinking without waiting for a meeting. The third is scale, where one person is in many rooms at once: digital avatars of the Chinese entrepreneur Luo Yonghao and his co-host ran a six-hour livestream in June 2025 that drew more than 13 million views and about $7.66 million in sales, TechNode reported. We cover the meeting use in our guide to an AI clone for meetings.
The record also shows what goes wrong. Kelly Monahan, a former Upwork Institute executive, watched her clone start stuttering and repeating lines in front of about 200 hotel executives in 2025, and it had to be shut down mid-presentation, according to WSJ reporting carried by DiarioBitcoin. When Sidhu revealed his clone, a KBW analyst said she was still recovering from the shock, and his finance chief stressed that his own section “was me the whole time,” The Philadelphia Inquirer reported.
What almost none of these twins was built for is the moment Nvidia faced on November 12. Nothing in the public record says what happens to them when the executive leaves, retires or dies. The WSJ-based coverage named control of a clone’s data after its owner leaves as an open question, Expansión reported. We compare the tools in our guides to the best AI clone app and the best digital twin app.
Jensen Huang’s 33 years against the average CEO
Huang ran Nvidia for about four times as long as the typical departing chief executive, using averages from Spencer Stuart and Russell Reynolds. In our view, the longer a founder stays, the more of the company’s judgment lives in one head.
Averages are for chief executives who left their posts in the year shown. Sources: Nvidia; Spencer Stuart; Russell Reynolds Associates.
Succession and key-person risk: what the data says
Nvidia was extreme, but far from alone in lacking a successor who was ready. In the National Association of Corporate Directors’ 2019 survey, only 23% of boards had a formal written CEO succession plan, down from 36% in 2017, though 74% of directors had identified an emergency interim chief executive, NACD reported. A Stanford and Heidrick & Struggles survey found that more than half of companies could not immediately name a successor, that 39% had no viable internal candidate, and that boards spent about two hours a year on the subject.
Chief executives leave often, and founders are no exception. Challenger, Gray & Christmas counted 2,032 U.S. chief executive departures in 2025, a record 446 of them at public companies, its year-end report shows, and through July 2026 114 departing CEOs were founders of the companies they led. About one in seven successions in 2024 and 2025 was forced, according to The Conference Board.
The cost of losing a founder is real because the value of a founder is real. Bain & Company found that founder-led S&P 500 companies delivered 3.1 times the shareholder returns of the others from 1990 to 2014, its study says, and its 2026 update puts the advantage since 2015 at 2.1 times. An Ohio State study of 2,327 large U.S. companies found that founder-led firms beat the rest by 4.4% a year after controls, the university reported.
Boards have three standard tools for key-person risk. Key-person insurance pays the business when an essential person dies, usually five to 10 times the person’s pay, Policygenius says. A continuity plan settles who runs and owns the business, and a succession plan names the next leader, which we compare in the best succession planning tools. In our view, none of the three keeps the founder’s judgment. That is the gap Huang tried to close by reasoning out loud, and the one an AI twin built from the founder’s own words is designed to fill, as we explain in how to capture founder knowledge.
How ready boards say they are for a CEO to leave
Share of boards or companies in each survey. Most have someone in mind for an emergency, and far fewer have a written plan.
Sources: National Association of Corporate Directors, Governance Challenges 2019; Stanford Graduate School of Business and Heidrick & Struggles, 2010 CEO Succession Planning Survey. The surveys asked different groups in different years.
Deepfake CEOs are already stealing millions
If a founder does not control a copy of himself, someone else may make one. In early 2024 an employee in Arup’s Hong Kong office joined a video call in which the chief financial officer and several colleagues were deepfakes, then made 15 transfers totaling HK$200 million, about $25.6 million, Fortune reported. Deloitte uses the case when it projects that generative AI could push U.S. fraud losses from $12.3 billion in 2023 to $40 billion by 2027, its analysts wrote.
In early November 2026, federal prosecutors in Manhattan charged three men with stealing at least $129 million between about July 2024 and February 2026 by impersonating chief executives, Gizmodo reported. Prosecutors say emails posing as a company’s chief executive were followed by calls in which AI copied his voice and at times his face, and then by fake lawyers who gave the wire instructions, often for a “secret acquisition,” GovInfoSecurity reported. The charges are allegations that have not been tested in court.
Some attacks failed, and the failures are instructive. Fraudsters set up a Microsoft Teams meeting with a senior WPP executive using a clone of chief executive Mark Read’s voice and YouTube footage of him, and Read warned staff that “just because the account has my photo doesn’t mean it’s me,” Marketing Beat reported. In July 2024 a Ferrari executive took a call from a voice clone of chief executive Benedetto Vigna and asked which book Vigna had recommended days earlier, at which point the caller hung up, MIT Sloan Management Review reported. In the Manhattan case, the back office of a New York investment bank stopped a $2.85 million wire.
The numbers are now large enough that regulators publish them. The FBI’s 2025 Internet Crime Report said 22,364 complaints mentioned AI, with adjusted losses of $893 million, and that fraud rings use “AI-generated videos and voices of celebrities, CEOs, or trusted figures.” The U.S. Treasury’s FinCEN issued an alert on deepfake fraud in November 2024, and Singapore’s police and central bank warned in March 2025 about video calls with deepfaked “senior executives.” In our view, the lesson for founders is that a familiar face and voice prove nothing, and a founder who decides deliberately how a copy of himself may be used is far safer than one who leaves it to whoever builds one first.
What deepfaked executives have taken
Reported losses in millions of dollars. Arup’s loss came from a single video call. The others are victims named in the federal case filed in Manhattan in November 2026, as described by Gizmodo; the charges are allegations.
Sources: Fortune (Arup, May 2024); Gizmodo, citing the U.S. Attorney’s charging papers, November 2026. Amounts are approximate.
What the law says about digital replicas
The law is moving toward one principle: a person’s face and voice belong to that person, and a copy needs consent. In the U.S. Senate, the NO FAKES Act of 2026 would create a federal right in voice and visual likeness that could be licensed during life but never sold, with licenses for living adults capped at 10 years. After death the right would pass by will and last at least 10 and no more than 70 years. The Senate Judiciary Committee passed it unanimously on June 18, 2026, with proposed damages of up to $750,000 for an online service that fails to comply, Music Business Worldwide reported, but it still needs both chambers and the President’s signature.
The states and Europe moved first. Tennessee, California and New York have each written digital replicas into law, New York’s 2025 statute requires the consent of heirs or executors for commercial use of a dead person’s likeness for 40 years, and the European Union has required deepfakes to be disclosed since August 2, 2026. Denmark introduced a bill on October 8, 2026, that would ban sharing realistic deepfakes of a person’s face or voice without consent, Euronews reported, though it is not yet law.
The laws on digital replicas, at a glance
Where the rules on AI copies of a person’s face and voice stood on November 14, 2026. This is a summary, not legal advice.
United States · S.4591
NO FAKES Act of 2026
- Passed Senate Judiciary
- Not yet law
Read the billA federal right in voice and likeness that can be licensed for up to 10 years but never sold during life, and passes by will after death for up to 70 years.
Tennessee · since July 2024
ELVIS Act
- Voice added to publicity rights
Read the summaryProtects a person’s voice, real or simulated, and makes liable anyone who distributes a tool whose primary purpose is copying an identifiable person without permission.
California · since January 2025
AB 2602 and AB 1836
- Contracts
- After death
Read the summaryReplica clauses need a description of uses and representation, and a dead performer’s digital replica needs the estate’s consent, with damages of at least $10,000.
New York · signed December 2025
S.8391 and S.8420
- 40 years after death
- Ad disclosure
Read the summaryHeirs or executors must consent to commercial use of a dead person’s likeness for 40 years, and ads must disclose AI “synthetic performers.”
Federal court, New York · July 2025
Lehrman and Sage v. Lovo
- Voice cloning claims proceed
Read the analysisTwo voice actors may pursue publicity, contract and consumer claims against an AI company they say cloned their voices.
European Union · since August 2026
AI Act, Article 50
- Disclosure
- Fines up to 3%
Read the guidanceAnyone deploying a system that generates deepfakes must disclose that the content is artificial, with a clear visual label or a spoken disclaimer for audio.
Who controls a founder’s AI twin after the founder is gone?
Who controls a founder’s AI twin after the founder is gone is the question the law has not answered. We could find no court ruling or statute that settles who owns a chief executive’s AI twin when the executive leaves or dies. The employment lawyer Kelly Dobbs Bunting of Greenberg Traurig told SHRM that work-for-hire clauses will expand to cover AI personas and that “it’s going to take lawsuits and legislative action to define these boundaries.” Gartner’s Daryl Plummer predicted in the same article that licensing clauses for AI versions of employees would appear in 70% of new employment contracts within three years.
ArentFox Schiff advises getting written consent for any avatar of a real person, and warns that an avatar generated by AI without meaningful human authorship may not be protected by copyright at all, so a company may not own a twin the way it owns other intellectual property. Under the NO FAKES Act as written, a person could license a replica of himself but not sell the right during life, and after death it would pass under his will. The direction of the law points toward the founder and his estate, and away from whoever happens to hold the files.
For a founder, the practical conclusion is to decide while you can, in writing and inside the system that holds the twin. Decide who may talk to your twin now, what it may say, and who takes control after your death. We cover ownership in do you own your AI clone and the build in how to make an AI version of yourself.
This is the problem Timeless AI™ was built around. Timeless AI™ builds an AI twin of you from the way you think, work and talk, so it answers with your reasoning as well as your phrasing, in your own voice, and you decide who it talks to. Executor Lock™ decides who controls your twin after your death and passes it to the executor you name, so your rules keep applying when you are no longer there to enforce them. For a founder, that means the people you trust can still ask how you would approach a decision and hear the answer in your own voice, on your terms. See how a twin works for a business.
How a founder should build a digital twin of themselves
Nvidia had the best digital Jensens in the world and still lost its founder’s judgment in a day. Six steps close that gap for any founder.
- 1
Decide what the twin is for
Build a twin for your businessAnswering staff questions, briefing a successor and keeping your reasoning for your family are different jobs. Choose the first one before you record anything.
- 2
Capture how you decide
How to capture founder knowledgeHuang reasoned out loud in every meeting. Most founders do not have 60 senior people in the room, so record your thinking on pricing, hiring and risk where your twin can learn it.
- 3
Name a successor and insure the key person
Key-person insurance explainedA twin keeps your judgment, and a person still has to run the company. Name an emergency successor and put key-person cover in place.
- 4
Set who may talk to it
How a Timeless twin worksDecide which people can ask your twin questions, from your leadership team to your family, and read what it tells them.
- 5
Decide who controls it after you
How Executor Lock™ worksExecutor Lock™ passes control of your twin to the executor you name, so your rules outlast you.
- 6
Build your twin in your own voice
Start your twin freeTimeless AI™ builds an AI twin from the way you think, work and talk, in your own voice. Free to start, no card.
Frequently asked questions
Short answers to the questions people ask about Jensen Huang’s digital twins and AI twins for CEOs and founders.
Is Jensen Huang coming back to Nvidia?
Nothing is decided. The Financial Times reported on November 15, 2026, that Huang has discussed a return as CEO with confidants, and Elon Musk and two large shareholders want the option kept open. Only Nvidia’s board can reinstate him, and it has said nothing about doing so.
Does Jensen Huang have an AI twin?
Not one that copies his judgment, as far as the public record shows. Nvidia has built several digital versions of him: a computer-generated Huang for 14 seconds of its April 2021 keynote, the talking avatar Toy Jensen in November 2021, and an avatar that narrated parts of his Computex 2024 keynote. In March 2026, Huang said his inbox and everything he has said had been “collected and becoming my AI.”
What is Toy Jensen?
Toy Jensen is a small animated avatar of Jensen Huang that Nvidia showed at GTC in November 2021. It used Nvidia’s Riva speech software and its Megatron 530B language model to answer engineers’ questions live.
Was Jensen Huang computer-generated in Nvidia’s 2021 keynote?
For 14 seconds, yes. In the April 2021 GTC keynote, the stretch from 1:02:41 to 1:02:55 showed a digital clone of Huang in a computer-generated copy of his kitchen, built from a full face and body scan. Nvidia revealed it in August 2021.
Did Nvidia use an AI avatar of Jensen Huang at Computex 2024?
Yes. Synthesia says it built an avatar of Huang that narrated several chapters of his Computex keynote on June 2, 2024, the day Nvidia made its ACE digital-human software generally available.
What was the Jensen Huang deepfake scam?
During Nvidia’s GTC keynote on October 28, 2025, a YouTube channel called “NVIDIA LIVE” streamed a deepfake Huang promoting a crypto scheme. PCWorld reported that it peaked near 100,000 live viewers, more than eight times the audience of Nvidia’s real stream.
What did Jensen Huang say about “my AI”?
On the Lex Fridman Podcast in March 2026 he said: “Take all my inbox, take everything that I’ve done, everything I’ve said. You know, it’s been collected and becoming my AI.” He was describing a far-future idea, not a product.
Did Jensen Huang resign from Nvidia?
Yes. Nvidia announced on November 12, 2026, that Huang had resigned as president, CEO and director, effective immediately. Chief financial officer Colette Kress is interim CEO while the board searches for a permanent successor.
Can an AI replace a CEO?
In our view, no. A chief executive is accountable to a board, shareholders and the law, and only a person can hold the office. An AI twin can take over the parts of the job that repeat what the CEO already thinks, such as answering staff questions and briefing new leaders.
Which CEOs use AI twins or avatars?
Zoom’s Eric Yuan, Klarna’s Sebastian Siemiatkowski and Customers Bank’s Sam Sidhu have used AI avatars to present results, and Reid Hoffman’s twin has given more than 75 presentations. Meta is building an AI Mark Zuckerberg for employees, and Uber teams rehearse pitches on a “Dara AI.”
What is the difference between an AI avatar and an AI twin?
An avatar is a likeness that delivers a script. A twin is built from a person’s own words and decisions so it can explain how they reason and answer questions they never scripted, and its owner decides who may use it.
Is it legal to make an AI clone of a CEO?
With the person’s consent and within the agreed uses, generally yes. Without consent it can break right-of-publicity laws in states such as Tennessee and New York, and the European Union requires deepfakes to be disclosed. The federal NO FAKES Act is not yet law.
Who owns a CEO’s AI twin after they leave or die?
No court or statute has settled it. The proposed NO FAKES Act would let a person license but not sell their replica rights during life, with the rights passing by will after death. The safest course is for the founder to decide in writing, which is what Executor Lock™ does for a Timeless AI™ twin.
How do deepfake CEO scams work?
A fraudster emails staff posing as the chief executive, then follows up with a call or video meeting in which AI copies the executive’s voice or face and pushes for an urgent, confidential payment. Prosecutors say one such scheme took at least $129 million between 2024 and 2026.
What is the NO FAKES Act?
Senate bill S.4591 would create a federal right in a person’s voice and likeness that can be licensed but not sold during life and passes by will after death for up to 70 years. The Senate Judiciary Committee passed it on June 18, 2026.
How do I make a digital twin of myself?
Decide what the twin is for, record how you think and decide in your own words, choose who may talk to it, and decide who controls it after you. Timeless AI™ builds an AI twin from the way you think, work and talk, in your own voice, and is free to start.
What is key-person risk?
It is the danger that a business depends on one person who cannot easily be replaced. Nvidia’s own annual report warned that it was “highly dependent on the services of our longstanding executive team.”
How many companies have a CEO succession plan?
In the National Association of Corporate Directors’ 2019 survey, only 23% of boards had a formal written CEO succession plan, although 74% had identified an emergency interim chief executive.
Sources
NVIDIA, NVIDIA Announces CEO Transition, November 12, 2026.
NVIDIA Corporation, Form 8-K exhibit, NVIDIA Announces CEO Transition, November 12, 2026.
NVIDIA Corporation, Annual Report on Form 10-K for fiscal 2026, filed February 25, 2026.
Bloomberg, Nvidia Lacks Clear Successor for Superstar CEO Who Built Company, January 27, 2026.
Lex Fridman Podcast #494, Jensen Huang transcript, March 23, 2026.
BusinessToday, “I can be fired”: Jensen Huang’s remarks on Nvidia succession resurface, November 12, 2026.
Wccftech, NVIDIA Has a Single Point of Failure, and That Is Jensen Himself, January 27, 2026.
NVIDIA Blog, From Our Kitchen to Yours: NVIDIA Omniverse Changes the Way Industries Collaborate, August 2021.
VICE, Nvidia reveals its CEO was computer generated in keynote speech, August 12, 2021.
NVIDIA Blog, NVIDIA CEO on accelerated computing, AI, Omniverse, avatars and robots at GTC, November 9, 2021.
NVIDIA Blog, How Omniverse Wove a Real CEO, and His Toy Counterpart, Together With Stunning Demos at GTC, December 21, 2021.
NVIDIA Newsroom, NVIDIA Releases Digital Human Microservices, June 2, 2024.
Synthesia, Synthesia debuts AI avatar of NVIDIA CEO Jensen Huang at Computex 2024, June 2024.
NVIDIA Blog, “Everybody Will Have an AI Assistant,” NVIDIA CEO Tells SIGGRAPH Audience, July 29, 2024.
NVIDIA Blog, New NVIDIA Digital Human Technologies Enhance Customer Interactions Across Industries, July 29, 2024.
Cryptopolitan, Huang sees Nvidia with 100M AI assistants, October 2024.
Fortune, Nvidia CEO Jensen Huang says IT will become the HR of AI agents, January 9, 2025.
TechCrunch, Nvidia’s AI avatar sat on my computer screen, January 9, 2025.
PCWorld, A deepfake video of Nvidia’s CEO sent viewers to a crypto scam, October 29, 2025.
Engadget, More people watched a fake NVIDIA livestream than the real thing, October 29, 2025.
Elon Musk, post on X, “That would be an honor”, November 11, 2026.
Elon Musk, post on X, petition to reinstate Huang, November 13, 2026.
Elon Musk, post on X, “a handshake with Jensen, not a written contract”, November 14, 2026.
Reuters via Yahoo Finance, Jensen Huang discusses comeback as Nvidia CEO with allies, FT reports, November 15, 2026.
Stand with Jensen, An Open Letter to the NVIDIA Board, November 2026.
CNBC, Elon Musk says it would be an “honor” to hire Nvidia’s Jensen Huang, November 12, 2026.
Tech Startups, Elon Musk backs petition to reinstate Jensen Huang as Nvidia CEO, November 13, 2026.
Fortune, CEO of Customers Bank Sam Sidhu AI clone led earnings call, April 28, 2026.
The Philadelphia Inquirer, Customers Bancorp AI investors call, April 28, 2026.
The Irish Times (Financial Times), Meta builds AI version of Mark Zuckerberg to interact with staff, April 13, 2026.
TechCrunch, After Klarna, Zoom’s CEO also uses an AI avatar on quarterly call, May 22, 2025.
Entrepreneur, Klarna uses an AI clone of its CEO to summarize earnings, May 22, 2025.
Popular Science, Watch a tech billionaire talk to his AI-generated clone, May 1, 2024.
Expansión, citing The Wall Street Journal, Jefes clonados por IA, May 26, 2026.
DiarioBitcoin, citing The Wall Street Journal, Los ejecutivos ya están creando clones IA, May 25, 2026.
HC Magazine, Uber staff are using an AI clone of their CEO to practice pitches, March 3, 2026.
JD.com Corporate Blog, JD.com debuts AI digital representative of founder Richard Liu, April 22, 2024.
Yicai Global, Digital incarnation of JD.com founder costs one-tenth of human hosts, April 23, 2024.
TechNode, Luo Yonghao’s digital avatar draws over 13 million viewers, June 17, 2025.
Fortune, UBS is seeing strong demand for AI-generated video avatars of its analysts, May 20, 2025.
The Register, AI boss bots not welcome, March 23, 2026.
IMD, Can leadership be delegated to a digital self?, September 21, 2026.
NACD, Governance Challenges 2019: CEO Succession Planning, 2019.
Stanford GSB and Heidrick & Struggles, 2010 CEO Succession Planning Survey, June 2010.
Challenger, Gray & Christmas, December 2025 CEO Turnover Report, February 4, 2026.
Challenger, Gray & Christmas, July 2026 CEO Turnover Report, August 31, 2026.
Russell Reynolds Associates, Global CEO Turnover Index, 2026.
Spencer Stuart, 2025 S&P 1500 CEO Transitions, February 2026.
Harvard Law School Forum on Corporate Governance (The Conference Board), Forced CEO Departures, September 19, 2026.
Bain & Company, Barriers and Pathways to Sustainable Growth: Harnessing the Power of the Founder’s Mentality, July 19, 2016.
Bain & Company, The Magic of Founder-Led Companies, February 9, 2026.
Ohio State News, Companies led by their founders are winners on the stock market, August 6, 2006.
Policygenius, Key person insurance, September 15, 2026.
Fortune, Arup revealed as victim of $25 million deepfake scam, May 17, 2024.
Gizmodo, DOJ says deepfaked CEOs and lawyers helped steal $129 million, November 7, 2026.
GovInfoSecurity, Deepfake CEO impersonation scam steals $129M from companies, November 9, 2026.
FBI Internet Crime Complaint Center, 2025 Internet Crime Report, April 2026.
Marketing Beat, WPP CEO’s identity usurped by deepfake scammers, May 13, 2024.
MIT Sloan Management Review, How Ferrari Hit the Brakes on a Deepfake CEO, January 27, 2025.
FinCEN, Alert on Fraud Schemes Involving Deepfake Media Targeting Financial Institutions, November 13, 2024.
Singapore Police Force, Joint Advisory on Scams Involving Digital Manipulation, March 12, 2025.
Deloitte Insights, Generative AI is expected to magnify the risk of deepfakes and other fraud in banking, May 29, 2024.
GovInfo, S.4591, NO FAKES Act of 2026, May 20, 2026.
Music Business Worldwide, NO FAKES: Senate panel backs bill, June 18, 2026.
Davis Wright Tremaine, Tennessee’s ELVIS Act, April 2024.
Manatt, California Enacts a Suite of New AI and Digital Replica Laws, September 2024.
Davis Wright Tremaine, Recent New York laws on AI in content and advertising, January 2026.
Skadden, New York Court Tackles the Legality of AI Voice Cloning, July 17, 2025.
Faegre Drinker, Commission confirms Transparency Code of Practice and publishes final guidelines, July 30, 2026.
Euronews, Denmark introduces a bill protecting people’s faces and voices from deepfakes, October 8, 2026.
SHRM, Protecting Digital Personas, December 16, 2024.
ArentFox Schiff, The Business of AI Avatars: Key Legal Risks and Best Practices, December 22, 2025.