원문 보기
November seventeenth, 2023. Around noon, Pacific time. Sam Altman is in a hotel room in Las Vegas. He's in town for a Formula One race. The night before, Ilya Sutskever texted him. Sutskever is OpenAI's chief scientist and co-founder. He wants to talk at noon. Altman joins a Google Meet. Four members of his board are on screen.
Sutskever speaks. Altman is fired. The announcement goes out within minutes. The board gives one reason. Altman, it says, was "not consistently candid in his communications." The face of ChatGPT. Fired on a video call. Five days later, he was back. Most of the people who fired him were not. Now this story is a film.
Andrew Garfield plays Sam Altman. And the studio that greenlit it walked away — after its parent company agreed to invest fifty billion dollars in OpenAI. This is not a story about a chatbot. It's a story about who controls it. Summer 2015. Elon Musk is turning forty-four. Among the guests at his birthday party is an old friend.
Larry Page, co-founder of Google. Late in the evening, the conversation turns to artificial intelligence. Musk is worried. What happens if machines become smarter than people? What if they replace us? Page, by Musk's account, isn't worried at all. If digital minds are the next step, then so be it. Musk pushes back.
He's on the side of humans. And Page calls him a "speciesist." Someone who puts his own species first. Musk would later call that the last straw. He'd tell the story for years — on podcasts, in interviews, and eventually, under oath. But the insult wasn't the real problem. The real problem was a lab in London.
A year earlier, Google had bought DeepMind. The most ambitious AI lab in the world. The reported price: around half a billion dollars. Musk had been an early investor in DeepMind. He tried to stop the sale. He failed. So now, the most advanced AI on the planet belonged to a man who thought Musk's concern for humanity was a kind of bias.
Enter Sam Altman. Thirty years old. President of Y Combinator, the most famous startup accelerator in Silicon Valley. In May 2015, he emails Musk with an idea. A nonprofit AI lab. A counterweight to Google. That summer, a small group meets for dinner at the Rosewood hotel in Menlo Park. A private room.
A few minutes from Sand Hill Road, where venture capital lives. At the table: Musk. Altman. Greg Brockman, an engineer who had just left his job as CTO of Stripe. And a handful of AI researchers. The question is simple. Is it too late to catch Google? Everyone agrees on one name. Ilya Sutskever. Born in the Soviet Union.
Raised in Israel. Trained in Toronto under Geoffrey Hinton, the man now called the godfather of AI. In 2012, Sutskever helped build AlexNet — a neural network that taught computers to recognize images, and kicked off the modern AI boom. Google bought their tiny startup. By 2015, Sutskever worked at Google Brain.
Google fought to keep him. Musk fought harder. He would later call Sutskever the linchpin of the whole project. December eleventh, 2015. OpenAI is announced. A nonprofit. Altman and Musk as co-chairs. Sutskever as research director. Brockman as CTO. The backers: Musk. Altman. Peter Thiel. Y Combinator's research arm.
Infosys. And one more name on the list. Amazon Web Services. Remember that one. The founding post says OpenAI will work "unconstrained by a need to generate financial return." The budget: one billion dollars. Pledged. Not paid. Pledged. Within three years, the biggest backer would walk out the door.
And the gap between those two words would shape everything that came next. By 2017, the people at OpenAI had figured something out. Ideas were cheap. Compute was not. To compete with Google, you needed enormous computing power. And computing power cost money. Far more money than donations could ever cover.
So the founders started asking an awkward question. Should OpenAI stay a nonprofit at all? Musk had an answer. According to emails OpenAI later published, he wanted majority control. He wanted to be CEO. And at one point he backed a plan to attach OpenAI to Tesla — with the car company as its cash cow.
Sutskever and Brockman pushed back. In a long email, they reminded Musk of his own fear. He worried that the head of DeepMind could become an AI dictator. Fine, they wrote. Then why build a structure where Musk could become one? In the same email, they turned to Altman. They admitted they didn't fully trust his judgment either.
They couldn't figure out what he really wanted. Six years later, Sutskever would act on that doubt. Musk left OpenAI's board in February 2018. The official reason was a possible conflict of interest with Tesla's own AI work. But he kept writing. And in December 2018, he put OpenAI's chances of staying relevant against Google, without a dramatic change, at zero.
"Not one percent." The billion dollars that had been pledged? Musk ultimately put in around thirty-eight million. Now the money problem was urgent. And it landed on Sam Altman. First, though, a small marketing lesson. In February 2019, OpenAI announced a language model called GPT-2. And said it was too dangerous to release in full.
Critics called it a stunt. Later that year, the full version came out. The world survived. In March 2019, Altman stepped away from Y Combinator. Reports would later say he was pushed out. YC co-founder Paul Graham rejected that word. Either way, Altman now had one job. OpenAI. That same month, OpenAI unveiled a new structure.
A "capped-profit" company, controlled by the nonprofit. Investors could now earn returns. But only up to a hundred times their money. Put in ten million dollars, and you could take out a billion. No more. A real sacrifice. Four months later, Microsoft invested one billion dollars. In return, OpenAI would run on Microsoft's cloud.
A lab founded to counter one tech giant now had another as its main partner. There was one more detail in the fine print. The nonprofit board stayed on top. It owned no shares. It answered to no investors. And it had the power to fire the CEO. At the time, nobody thought much of it. With Microsoft's money, OpenAI went all in on one idea.
Scale. Take a language model. Feed it more text. Give it more compute. See what happens. June 2020. GPT-3. One hundred seventy-five billion parameters. It could write essays, code, and poetry of questionable quality. Developers were impressed. Everyone else barely noticed. Inside, not everyone liked the direction.
A group of researchers felt the company was racing to ship products faster than it was learning to make them safe. The most senior of them was Dario Amodei, OpenAI's vice president of research. At the end of 2020, he left. His sister Daniela left too, along with several colleagues. In 2021, they founded a rival lab.
Anthropic. The first real split at OpenAI was not about money. It was about speed. November 2022. OpenAI has a newer model and a nagging fear: a competitor might ship a chatbot first. So a small team wraps the model in a simple chat window, and ships it. Internally, it's framed as a "low-key research preview." So low-key that, according to The Atlantic, some employees didn't even realize it had launched.
November thirtieth, 2022. ChatGPT goes live. No ad campaign. A blog post and a few tweets. Five days later, Altman posts an update. One million users. By January, analysts at UBS estimate a hundred million monthly users. At the time, the fastest-growing consumer app on record. The low-key preview had become the most talked-about product on the planet.
January 2023. Microsoft announces a new multiyear, multibillion-dollar investment. Reports put it at around ten billion dollars. And Sam Altman becomes the face of artificial intelligence. He testifies before the US Senate. He goes on a world tour. He meets presidents and prime ministers. And he asks governments to regulate his own industry.
While he's on the road, something is going wrong at home. The nonprofit board now has six members. Three work at OpenAI: Altman, Brockman, Sutskever. Three do not. Adam D'Angelo, the CEO of Quora. Tasha McCauley, a tech entrepreneur. And Helen Toner, an AI policy researcher. In October 2023, Toner co-authors a paper.
In one passage, it praises a competitor's approach to safety over OpenAI's. The competitor is Anthropic. Altman is not pleased. According to reports, he starts sounding out other board members about removing her. That turned out to be a mistake. Because the board members started comparing notes. Comparing notes is exactly what Ilya Sutskever had been doing.
For months, he'd been collecting concerns about Altman. Screenshots. Messages. Accounts from senior staff, including chief technology officer Mira Murati. Sutskever later testified that he put it all into a long memo for the independent directors. His summary of Altman, in his own words: "a consistent pattern of lying." The independent directors read it.
Then they did the math. Six seats. Altman and Brockman on one side. Sutskever, D'Angelo, McCauley and Toner on the other. Four votes. Enough. Brockman is not told. Friday. At noon, Altman is fired on the Google Meet call. Minutes later, Brockman is pulled into a separate call. He's removed as chair of the board, but asked to stay on as president.
That evening, he quits. Mira Murati becomes interim CEO. Microsoft — the company that had put billions into OpenAI — reportedly learns about the firing around a minute before the public does. The board explains nothing beyond its one sentence. Not to employees. Not to investors. Not to Microsoft. That silence would cost them everything.
Saturday. Investors start pushing to bring Altman back. On social media, employees reply to Altman's posts with heart emojis. One after another. The board begins talks with Altman about a possible return. Sunday. Altman walks into OpenAI's headquarters. As a visitor. He posts a photo of himself holding the guest badge, with a caption: "first and last time i ever wear one of these." The talks fail.
Late that night, the board names a new interim CEO. Not Murati. Emmett Shear, the former head of Twitch. That's three CEOs in three days. Hours later, Microsoft CEO Satya Nadella announces that Altman and Brockman will lead a new advanced AI research team. At Microsoft. Monday. Now the threat is clear.
If OpenAI's people follow Altman, Microsoft gets the talent. And the nonprofit keeps an empty building. Overnight, one phrase spreads across X. Mira Murati posts it. Then hundreds of employees. "OpenAI is nothing without its people." A letter starts circulating inside OpenAI. It demands the board resign and bring Altman back.
If not, the signers will quit and join him at Microsoft. By the end of the day, more than seven hundred employees have signed. Out of roughly seven hundred seventy. Among the names: Mira Murati. And Ilya Sutskever. The same morning, Sutskever posts a public message. "I deeply regret my participation in the board's actions." The man who fired Altman was now asking for him back.
Altman quotes the post. And adds three hearts. The remaining directors hold their ground. According to reports, Toner told executives that letting the company fall apart could still be consistent with the mission. Technically, she had a point. The nonprofit's charter put humanity ahead of the company.
But a charter doesn't pay salaries. And it doesn't own servers. Tuesday. Late that night, a deal. Altman returns as CEO. Toner and McCauley leave the board. Sutskever leaves the board. Adam D'Angelo stays, the only director to survive. The new chair is Bret Taylor, former co-CEO of Salesforce. Joining him is Larry Summers, former US Treasury Secretary.
Emmett Shear's tenure ends after roughly two days. On paper, the nonprofit board controlled OpenAI. It had the right to fire the CEO. And it used that right. In practice, power belonged to other people. Employees whose equity depended on the company. Microsoft, which owned the computing power. Investors who held the money.
The board had the power to fire him. It didn't have the power to make it stick. Officially, Sutskever kept his job as chief scientist. In practice, according to Business Insider, within weeks he hadn't been seen at the office, and his role was "to be determined." March 2024. An outside law firm finishes reviewing the firing.
Its finding: trust between Altman and the old board had broken down. But his conduct did not require his removal. Altman returns to the board. Case closed. Then came May. May thirteenth. OpenAI demos a new voice assistant. One of its voices, called Sky, sounds a lot like Scarlett Johansson. Who, in the film Her, played an AI that a lonely man falls in love with.
During the launch, Altman posts one word: "her." May fourteenth. Ilya Sutskever announces he's leaving OpenAI. Hours later, Jan Leike, who co-led the company's team for controlling future superintelligent AI, resigns too. Three days later, he explains why. Safety culture, he writes, had "taken a backseat to shiny products." Within days, that team is dissolved.
The same week, journalists at Vox report something else. Departing employees were asked to sign lifelong non-disparagement agreements. Refuse, and they could lose their vested equity. For many of them, millions of dollars. Altman says he didn't know. He calls himself "genuinely embarrassed." May twentieth.
Johansson reveals that Altman had asked her to voice the assistant. She had said no. By then, OpenAI had already paused the voice. It insists it was never meant to imitate her. Vox then publishes documents. They carry signatures of OpenAI's top executives. And then Helen Toner finally speaks. On a podcast, she gives the old board's side.
Among her claims: the board wasn't told about ChatGPT before launch. In her words, they "learned about ChatGPT on Twitter." June 2024. Sutskever announces a new company. Safe Superintelligence. One goal, one product, no distractions. No chatbot. Then September. Mira Murati — briefly OpenAI's interim CEO — announces she's leaving.
On the same day, the company's chief research officer and another senior research leader leave too. Count the founders. Of the eleven people named in December 2015, only a handful are still there. And yet, the business has never been bigger. The money keeps arriving. Valuations keep climbing. In October 2025, OpenAI completes its restructuring.
The for-profit arm becomes a public benefit corporation. The nonprofit is renamed the OpenAI Foundation, and keeps a stake of around a quarter of the company. Microsoft holds roughly the same. On paper, the nonprofit is still in charge. On paper, it was in charge in November 2023 too. By now, almost everyone who tried to stop Sam Altman has left.
Almost everyone. One man still wanted to unwind the whole thing. And he had a court date. Elon Musk had sued OpenAI in 2024. His claim: he funded a charity, and Altman turned it into a business. OpenAI's reply: Musk had wanted a for-profit too. As long as he ran it. In February 2025, Musk tried another route.
A group he led offered ninety-seven point four billion dollars to buy control of OpenAI. Altman's reply on X began with "no thank you." Then he offered to buy Twitter instead. For nine point seven four billion. One tenth. Nobody missed the joke. April 2026. A federal courthouse in Oakland, California.
Jury selection begins on April twenty-seventh. The judge is Yvonne Gonzalez Rogers. Musk testifies for three days. Sutskever testifies. So does Shivon Zilis, a former OpenAI board member and the mother of four of Musk's children. Musk's demands are not small. Return more than one hundred thirty billion dollars to the nonprofit.
Remove Altman and Brockman. Unwind the restructuring. Near the end of testimony, in the middle of a dispute between the lawyers, the judge says it out loud. "This entire trial is a giant irony." May eighteenth, 2026. After eleven days of testimony and arguments, the jury returns a unanimous verdict.
It doesn't rule on who betrayed whom. It rules that Musk waited too long to sue. The verdict is advisory. The judge says she agrees with it. Musk calls it a "terrible precedent." And promises to appeal. Meanwhile, the money hasn't slowed down. In February 2026, OpenAI announces a funding round anchored by three companies.
Nvidia, thirty billion dollars. SoftBank, thirty billion. Amazon, fifty billion. By the end of March, the round closes at one hundred twenty-two billion dollars. The valuation: eight hundred fifty-two billion. As part of the deal, OpenAI expands its commitments to Amazon's cloud. Up to a hundred billion dollars over eight years.
And in the same months, Amazon had something else on its hands. A film. Amazon MGM Studios had greenlit a movie about the five days in November. Director: Luca Guadagnino. Star: Andrew Garfield, as Sam Altman. A movie whose script, according to one industry newsletter, portrayed Altman as "a liar and a master schemer." Amazon dropped it.
In late June, the distributor Neon picked it up. Worldwide rights. An Oscar campaign. A US release on December twenty-fifth. OpenAI, meanwhile, prepares to go public. It files confidentially this summer. Then, in September, Altman tells Fortune that right now would be "an ill-advised moment to go public." His verdict on a listing this year: "not 2026." Possibly the biggest IPO in tech history.
It can wait. The movie can't. So let's add it up. In 2015, OpenAI was founded so that no single company would control the most powerful technology of our time. In 2026, it's one of the most valuable private companies in the world. Backed by Microsoft. Backed by Nvidia. Backed by Amazon. The board that tried to rein in its CEO lost.
The founder who sued lost. The chief scientist who fired Altman now runs his own lab. In July, Nvidia agreed to invest five billion dollars in it. And the one person who's still there, in charge, is the man the board said wasn't consistently candid. On December twenty-fifth, the film comes out. Andrew Garfield as Sam Altman.
Yura Borisov as Ilya Sutskever. Ike Barinholtz as Elon Musk. Mark Rylance as Geoffrey Hinton — the man who trained Sutskever, and who spent his later years warning about the very technology his students built. One more detail. Remember the list of backers from December 2015? Amazon Web Services was on it.
Eleven years later, Amazon wrote OpenAI's biggest check. And walked away from the movie about it. So here's the question. If a board can fire a CEO — and still lose — who actually controls the most powerful technology of our time? Tell me in the comments. Next time: the people who walked out of OpenAI in 2021. And built its biggest rival.
Comments 0