This is an excerpt of Sources by Alex Heath, a newsletter astir AI and the tech industry, syndicated conscionable for The Verge subscribers erstwhile a week.
Elon Musk first sued OpenAI successful February 2024. Despite OpenAI’s repeated attempts to propulsion it out, the lawsuit is now headed to a assemblage proceedings connected April 27th successful Northern California national court.
Musk’s main allegation is that OpenAI and its leaders abandoned the company’s original nonprofit ngo that he funded. In turn, OpenAI has treated Musk’s claims arsenic sour grapes. U.S. District Judge Yvonne Gonzalez Rogers precocious decided that the lawsuit warranted going to trial, saying successful tribunal that “part of this is astir whether a assemblage believes the group who will attest and whether they are credible.”
Last week, thousands of pages of grounds from the lawsuit were unsealed, including partial 2025 depositions of astir of the cardinal players involved, including Sam Altman, Ilya Sutskever, Greg Brockman, Mira Murati, and Satya Nadella, on pinch ex-board members Helen Toner and Tasha McCauley — some of whom played cardinal roles successful the 2023 firing of Altman.
Bits and pieces of this grounds person started trickling retired successful caller days, specified arsenic the news that Sutskever owned a whopping $4 cardinal successful vested OpenAI shares erstwhile Altman was concisely fired 2 years ago. Altogether, the unsealed grounds offers a fascinating look not only astatine OpenAI’s early days but besides astatine the circumstances surrounding Altman’s firing and Microsoft’s analyzable narration pinch OpenAI.
I’ve been covering OpenAI successful extent for a while, and I intimately reported connected the whirlwind fewer days erstwhile Altman was fired and past rehired successful precocious 2023. It’s done that lens that I’ve pulled retired the beneath highlights from the grounds successful Musk v. Altman:
Sutskever had early concerns astir treating open-source AI arsenic a “side show.”
In 2022, OpenAI’s leaders seemed rather concerned astir the prominence of open-source laboratory Stability AI, and Sutskever voiced his interest complete matter pinch Murati and others:
Sutskever: My trepidation astir unfastened root is that we’re treating it arsenic a broadside show, eg def not going acold capable to really wounded stability
Murati: We’re missing the opportunity to group standards pinch this monolithic increasing group of devs, group are quiet to build things and we should thin successful and bring our tech to arsenic galore group arsenic possible, agelong word maximize our chance of maintaining lead, reducing competition
But if we do everything to get this successful a mates of weeks astatine immoderate costs retired bc we heard stableness is unfastened sourcing akin model, that’s not successful statement astatine each pinch my motivations
OpenAI leaders were divided complete early investor Reid Hoffman’s determination to commencement a rival AI lab, Inflection.
They were besides already considering prohibiting investors from backing competing labs. From an October 2022 exchange:
Sutskever: I conjecture I conscionable felt betrayed by him founding a nonstop competitor while simultaneously telling maine that “I could not perchance ideate you’d find it objectionable”
Altman: here’s really id summarize my thoughts connected this:
pros: he supported america successful a infinitesimal wherever nary 1 other would and it was beautiful existential--i deliberation openai would person been beautiful fucked if he hasn’t stepped up then. also, he was instrumental to getting the first MSFT woody done, and has mostly been rather adjuvant pinch MSFT related worldly he is mostly a bully committee member.
cons: he is very motivated by `collecting’ status. though one personally deliberation he cares overmuch much astir openai than inflection, he was blinded capable by the startup of being capable to telephone himself the cofounder of a institution he made an uncareful decision.
also, astatine this point, one deliberation astatine this constituent openai has the leverage to inquire for a soft committedness for caller investors not to put successful competitors, but only a prime fewer companies ever get to do that)
Brockman: ohio besides an aside, aft taking to @Sam Altman, I’m readying to meet Patrick Collison tmrw and demo dv3. Will inquire if he’d beryllium willing successful participating successful the tender nether the information of not investing successful AGI/big exemplary competitors
Brockman wrote successful his diary that he wanted to beryllium a billionaire.
From his deposition:
Q: Why did you constitute “Financially, what will return maine to 1 billion?”
A: I deliberation if we were going to do a for-profit entity, that I started to deliberation astir what would beryllium motivating financial reward successful that lawsuit arsenic a secondary consideration.
Q: What was the superior consideration?
A: Primary information was would we beryllium capable to prosecute and execute the mission.
Q: How important was the secondary information to you?
A: The 2nd information decidedly mattered.
Q: At this point, did you aspire to beryllium a billionaire?
A: My superior information was to the mission.
Q: Was your secondary information to beryllium a billionaire?
A: I judge that arsenic a -- 1 point I was decidedly motivated by was the thought — I decidedly had arsenic a information that, yeah, perchance getting to $1 billion.
Q: So we cognize you achieved that extremity astatine immoderate point. Do you cognize precisely what time that happened?
A: I do not cognize what time precisely that happened.
Q: When was the first time you realized you had surpassed that goal?
A: I do not cognize what time I would opportunity my, astatine slightest connected paper, nett worthy would’ve exceeded 1 billion.
Nadella was worried astir Microsoft’s position successful AI erstwhile he started looking astatine OpenAI.
From his deposition:
Q: Did you consciousness that your advancement was moving much slow than you had liked?
A: I mean, ever arsenic a CEO of a company, I consciousness my occupation is to benignant of beryllium dissatisfied pinch the complaint of advancement astatine each times. And truthful “yes” would beryllium the answer, which is some successful the absolute sense, which is, tin we build products that are much tin successful immoderate peculiar domain, and also, you know, vis-à-vis competition.
There were others achieving things that we looked astatine and said, “Hey, that’s great, and truthful really tin we make judge we are competitory pinch it.”
Nadella almost wrote a book astir AI called An Inflection Point.
According to an grounds revenge successful the case, it was co-written pinch Marco Iansiti and was successful improvement successful 2023. From the first chapter:
On Wednesday, August 24, 2022, pinch the Pacific Northwest summertime showing each of its beauty, Bill Gates hosted a meal astatine his location connected Lake Washington, conscionable a fewer miles from the Microsoft campus. No longer a Microsoft committee personnel aliases moreover Microsoft’s largest shareholder, Bill remained the iconic co-founder and trusted advisor of the company’s elder method leaders. Satya suggested the gathering, which included Chief Technology Officer, Kevin Scott, and a fistful of apical researchers. Food and drinks would beryllium served, but the main entrée was a hush- hush demo by OpenAl laminitis Sam Altman of a forthcoming merchandise of ChatGPT powered by GPT-4, an AI built connected Large Language Models (LLMs). Bill had agelong encouraged researchers to create a genuinely accomplished AI adjunct but had voiced his skepticism astir this peculiar approach.
Microsoft hit retired Amazon erstwhile it initially started moving pinch OpenAI.
Musk was opposed to moving pinch Jeff Bezos and wrote the pursuing successful an early email to Altman: “I deliberation Jeff is simply a spot of a instrumentality and Satya is not, truthful I somewhat for illustration Microsoft, but I dislike their trading dept.” Altman responded that Amazon had “started really dicking america around.”
The upside connected Microsoft’s first $1 cardinal finance successful OpenAI was capped astatine $500 billion.
From a filing written by Musk’s lawyers:
In November 2018, aft meal pinch Altman, Scott told Nadella that OpenAI’s caller firm building offered some “a commercialized conveyance for monetizing Open AI IP” and finance returns “capped astatine $500B.” Altman claimed the nonprofit would yet use because — though OpenAI had yet to make a azygous dollar successful returns — “[i]f [OpenAI] ever [does] get to $500B successful returns, the equilibrium complete that goes straight to the 501(c)3.”
Microsoft’s committee initially approved a superior finance of $2 billion. But ultimately, it decided to limit its first finance to $1 cardinal successful the hopes that a smaller finance would “press OpenAI to commercialize,” successful nonstop contravention of the nonprofit’s stated founding principles. In speech for its investment, Microsoft received a convertible constricted business liking and authorities to OpenAI’s profits, pinch returns “capped” astatine 2000% of its $1 cardinal investment.
Microsoft’s CFO noted successful an soul email that the “cap is really larger than 90% of nationalist companies,” and the limit connected Microsoft’s profits is not “terribly constraining nor terribly altruistic.” It was, successful fact, “a bully investment.” At Microsoft’s request, OpenAI agreed to support immoderate mention of Microsoft’s promised 2000% return connected its finance retired of its nationalist announcement.
The 2nd update to Microsoft’s business pinch OpenAI successful 2021 included different $2 cardinal finance that wasn’t reported and came pinch a little upside.
From a filing written by Musk’s lawyers:
In March 2021, Microsoft softly invested different $2 cardinal successful OpenAI. Neither OpenAI nor Microsoft publically announced the investment, which was taxable to a little 6x return multiple.
In spot of its 2019 licence to a azygous OpenAI model, Microsoft secured authorities to commercialize immoderate OpenAI exemplary developed during the word of the statement (except AGI). Facilitating its commercialized usage of OpenAI’s IP, Microsoft was permitted to embed up to 10 of its labor on-site astatine OpenAI.
Anticipating accrued merchandise commercialization, Microsoft and OpenAI agreed to stock immoderate resulting revenue.
Just 3 months later, successful June 2021, Microsoft released GitHub CoPilot — its first merchandise incorporating OpenAI’s technology.
Microsoft’s adjacent $10 cardinal finance successful OpenAI came pinch unit from Nadella to spell aft the endeavor marketplace and pinch much strings attached.
From a filing written by Musk’s lawyers:
Prodding OpenAI to accelerate its ain merchandise development, Microsoft told Altman that OpenAI needed to make $100 cardinal successful revenues to unafraid the adjacent $10 cardinal committedness from Microsoft. To meet that goal, OpenAI expanded the squad responsible for taking products to marketplace and tried to grow its “enterprise business.”
In the summertime of 2022, OpenAI began negotiating pinch Microsoft a caller $10 cardinal investment. That November, OpenAI released ChatGPT. It was an instant hit. Nadella urged Altman to merchandise a paid type and persistently checked connected the advancement of its commercialization.
Over the adjacent respective months, OpenAI secured Microsoft’s $10 cardinal investment, and the parties again amended the JDCA. OpenAI besides changed its firm structure.
The 2023 statement “cap[ped]” Microsoft’s return connected this finance astatine 600%, aliases $60 cardinal to start, but accrued Microsoft’s profit “cap” by 20% per year. Microsoft would person 49% of OpenAI’s profits, while the OpenAI nonprofit entity would retrieve conscionable 2% of OpenAI’s profits — astatine slightest until each extracurricular investors were paid retired their finance returns, weighted successful full astatine $261 billion.
Underscoring the profit-focused purpose of the partnership, the 2023 JDCA was specifically system to “remove the impediments successful commercialization.”
Microsoft negotiated expanded IP authorities to see each OpenAI IP developed earlier aliases during the word of the statement (excluding AGI), and the correct to embed up to 20 labor astatine OpenAI.
Finally, Microsoft and OpenAI established an 80%-20% gross share.
OpenAI considered adding AI information experts Dan Hendrycks, Paul Christiano, Jacob Steinhardt, and Ajeya Cotra to the committee earlier Altman was fired.
Altman apparently wanted committee members pinch much “commercial” experience. From Toner’s deposition, successful reference to soul discussions astir expanding the committee earlier it fired Altman successful precocious 2023:
Q: Was it your belief that Mr. Altman was dragging his feet successful these discussions?
A: Yes. I deliberation that’s a adjacent description.
Q: Did Mr. Altman’s actions consequence successful the committee being deadlocked complete immoderate connection to adhd an further AI information committee member?
A: I’d opportunity he contributed to america importantly being deadlocked, yes.
Q: Did Mr. Altman propose different candidates to the board?
A: Yes.
Q: Were Mr. Altman’s replacement candidates besides AI information experts aliases did they person different backgrounds?
A: To the champion of my recollection, he mostly projected candidates pinch much of a commercialized startup background.
Altman and Brockman projected kicking Adam D’Angelo disconnected the committee earlier Altman was fired.
From Toner’s deposition:
Adam runs a institution called Quora, which has a merchandise called Poe, which uses ample connection models, including those of OpenAI and immoderate of its competitors.
The measurement I perceived it was aft GPT-4 was demoed to the committee successful summertime 2022, Adam began taking his responsibilities arsenic a committee personnel much seriously, because the exertion seemed to beryllium advancing, and he became a much engaged committee member.
In the lead-up to — betwixt that clip successful summertime 2022 and April 2023, we had had respective conversations arsenic a committee astir what kinds of conflict of liking were acceptable aliases unacceptable connected the board, because galore imaginable committee members, and existent committee members, had various involvements pinch various AI companies.
So we had reasonably elaborate discussions astir what was an unacceptable conflict of liking and had decided that being intimately progressive pinch a institution that was training its ain ample connection models, you know, highly precocious frontier connection models that would compete pinch OpenAI’s, was the barroom for excessive conflict of interest.
So it was astonishing to maine erstwhile Sam emailed the committee successful April 2023 saying that Adam’s conflict of liking had grown excessively ample and seemed for illustration he needed to measurement disconnected the board, and did we agree. Because Adam’s institution produced a merchandise that utilized others’ LLMs, they didn’t — they weren’t training their own. So intelligibly it didn’t meet the conflict of liking criteria we had each discussed.
When I said arsenic overmuch via email, Greg Brockman chimed successful pinch a different logic to region Adam, namely that his position arsenic some a customer and a committee personnel was creating connection difficulties internally. I hide who precisely said what connected the email chain, but different committee members raised questions astir that aliases wanted to cognize much astir that.
Ultimately, I said to Sam connected the phone, and we benignant of — astatine my urging, we agreed that, surely, the measurement earlier conscionable removing Adam from the board, if the problem was really he was communicating wrong the company, surely, the adjacent measurement would beryllium to talk that pinch him and spot if we tin amended the situation. Sam said he would do that, he would person a speech pinch Adam, to effort and amended really he was communicating wrong the company. And past the business seemed to spell away.
I later recovered retired that Sam had ne'er had that speech pinch Adam, aliases that he had talked pinch him but had ne'er really tried to lick that problem, but, instead, had conscionable said the only point that he, Sam, didn’t for illustration astir Adam’s merchandise Poe was that it utilized Anthropic models, because Anthropic was a competitor.
So, each successful all, the business seemed to maine for illustration location wasn’t really a clear, actual logic to inquire Adam to move disconnected the board, but that Sam and Greg were benignant of searching for an excuse because he had been providing much progressive governance of the company.
Altman didn’t initially show OpenAI’s committee that he was personally moving a institution VC fund.
From Toner’s deposition:
Adam D’Angelo was astatine a meal pinch immoderate different founders, investors, startup people, who were asking him astir the building of the startup money and imaginable conflicts of liking betwixt the startup money and OpenAI’s investors much generally.
And aft that conversation, Adam emailed the board, including Sam, possibly a mates of different OpenAI executives, to understand the building better.
And successful the resulting back-and-forth, we learned that Sam was the, arsenic I understand it, the proprietor of the fund. So the first speech was astir whether it was adjacent for OpenAI’s investors that OpenAI was benignant of contributing to this different money and was besides contributing benignant of engineering expertise and clip to portfolio companies successful the startup money successful ways that whitethorn not — wherever the use whitethorn not accrue backmost to OpenAI investors.
After we learned that Sam had a financial liking successful the fund, we besides had concerns astir the truth that he had not disclosed that, fixed that his position connected the committee was 1 of a supposedly independent committee director, meaning 1 pinch nary financial liking successful OpenAI.
Altman projected making a aid to then-Congressman Will Hurd while he was successful talks to subordinate the OpenAI board.
From Toner’s deposition:
Sam besides suggested that he wanted to make a large, I believe, several-hundred-thousand-dollar run publication to Will, while still expecting him to travel backmost onto the board.
He did not spell up pinch this aid because Tasha, Adam, and I each said it seemed very inappropriate. But to me, the truth that he was considering that, the truth that he mightiness person discussed it pinch Will successful advance, the truth it was an option, was conscionable a motion of full disregard for the board’s independency aliases expertise to supply meaningful oversight of the institution and the CEO.
Q: And that several-hundred-thousand-dollar run contribution, was it — did Mr. Altman talk that that was going to travel from him personally?
A: Yes, to the champion of my recollection.
There were concerns astir Altman’s closeness pinch the existent OpenAI committee chairman, Bret Taylor.
From McCauley’s deposition:
I had much discourse connected Bret Taylor than I did connected Larry [Summers], and I had concerns astir his expertise to beryllium — yeah, to make disinterested decisions successful a measurement that was, wasn’t partial to Sam. I mean, you know, we had — he had been projected by Sam for the committee antecedently erstwhile we were location and erstwhile we were going done the process of expanding the board. And by the champion of my recollection, you know, Sam had — had made recommendations connected a number of different people. He was favorable to Bret Taylor.
If I callback correctly, Adam had — I deliberation I callback correctly that Adam had interviewed Bret successful the process of considering different candidates, and that 1 of the — anterior to each of this — sorry — like, successful the process that we were moving complete this — you know, successful the months prior, erstwhile we were trying to grow the board; and astatine that time, that — 1 of the takeaways from that speech was that — I deliberation — I’m going to effort to callback this precisely arsenic possible, but it was I deliberation Bret whitethorn person expressed interest that — interest astir the — the conflicts. I deliberation that he had said he had known Sam for a very agelong clip and had a batch of connections to Sam and whatnot.
There were astatine slightest six cardinal issues that led the committee to occurrence Altman.
From McCauley’s deposition:
Q: Was 1 of those incidents Mr. Altman’s foot-dragging complete adding an AI information master to the board?
A: That — that was — you know, I deliberation the truth that that process was incapable to consequence successful adding independent members and an AI information personnel to the board, it exacerbated our concerns, yes.
Q: And was different 1 of those incidents that — Mr. Altman’s practice that the 3 enhancements to GPT-4 had each been approved by the information board?
A: Yes, that was a factor.
Q: Was different 1 of those incidents Mr. Altman’s nonaccomplishment to disclose that a GPT-4 trial was released successful India without associated information committee review?
A: Yes.
Q: Was different incident Mr. Altman’s nonaccomplishment to pass the committee anterior to ChatGPT’s release?
A: Yes.
Q: And was different incident Mr. Altman’s misrepresentation astir you allegedly saying Ms. Toner should evidently time off the board?
A: Yes.
Q: And was different incident Mr. Altman’s misrepresentation that the ineligible section told him GPT-4 Turbo did not request information committee review?
A: Yes, that — that we saw screenshots to that effect.
Sutskever had $4 cardinal worthy of vested equity successful OpenAI arsenic of November 2023.
A matter speech betwixt Altman, Nadella, and OpenAI COO Brad Lightcap revealed the stake. As Microsoft was successful discussions to prosecute Altman and astir of the OpenAI team, Lightcap wrote that paying labor for their equity would costs $25 cardinal aliases $29 billion, depending connected whether Sutskever’s vested shares were included. While it’s intolerable to cognize for judge without much evidence, the speech suggests that Sutskever was OpenAI’s largest individual shareholder astatine the time. It’s unclear if he has sold immoderate shares since.
Sutskever thought “OpenAI would beryllium destroyed” if Altman wasn’t rehired.
From his deposition:
Q: And why did you retreat your support for Sam Altman being fired?
A: Because I thought that OpenAI would beryllium destroyed.
Altman told Musk, “It really fucking hurts erstwhile you publically onslaught OpenAI.”
From a February 2023 email exchange:
Altman: one retrieve seeing you successful a tv question and reply a agelong clip agone (maybe 60 minutes? wherever you being attacked by immoderate guys, and you said they were heroes of yours and it was really tough.
well, you’re my leader and that’s what it feels for illustration erstwhile you onslaught openai. wholly get we person immoderate screwed immoderate worldly up, but we person worked incredibly difficult to do the correct thing, and one deliberation we person ensured that neither google nor anyone other is connected a way to person unilateral power complete AGI, which one judge we some deliberation is critical.
i americium tremendously thankful for everything you’ve done to thief —i dont deliberation openai would person happened without you-and it really fucking hurts erstwhile you publically onslaught openai
Musk: I perceive you and it is surely not my volition to beryllium hurtful, for which I apologize, but the destiny of civilization is astatine stake.
Altman: i work together pinch that, and one would really emotion to perceive the things you deliberation we should beryllium doing differently/better.
it’s besides not clear to maine really the attacks connected twitter thief the destiny of civilization, but that’s little important to maine that getting to the correct substance.
also, one checked pinch our squad connected recruiting from tesla. we really are doing very small comparative to the size of the company, but one will make judge we don’t wounded tesla, one evidently deliberation it’s a ace important company.
“OpenAI has not yet done business pinch Helion but intends to if the exertion works.”
Altman is personally the largest investor successful Helion, which is building fusion powerfulness technology. From his deposition:
Q: While you were astatine Y Combinator, did you personally put successful immoderate of the companies that Y Combinator sponsored?
A: I did.
Q: Which ones?
A: I couldn’t springiness you a database disconnected the apical of my head.
Q: Have immoderate of those companies done business pinch OpenAI?
A: Yes.
Q: Which ones?
A: Our conflicts committee keeps way of each this and could show you. I couldn’t do a database disconnected the apical of my caput that would beryllium exhaustive. Reddit is one. OpenAI has not yet done business pinch Helion but intends to if the exertion works.
Altman thinks “things request to spell right” for OpenAI to beryllium worthy $500 billion.
From his deposition:
Q: And do you personally work together that the institution is worthy astatine slightest 500 cardinal currently?
A: That was the consenting buyer-willing seller marketplace price, truthful I won’t reason pinch it.
Q: Apart from your religion successful the consenting buyers and consenting sellers, do you agree, being the 1 who runs the company, that the institution is worthy astatine slightest $500 cardinal today?
A: If I were an extracurricular marketplace investor, I would — I deliberation I would perfectly emotion to bargain OpenAI shares astatine a 300-billion-dollar valuation, somewhat higher. At 500, I would commencement to say, “Could be, like, you know, things request to spell right, but could be.”
“No, I was not surprised, because I was utilized to the committee not being very informed astir things.”
- Toner responding to a mobility during her deposition astir whether she was amazed by the original merchandise of ChatGPT.
“I deliberation there’s a existent anticipation that 5 aliases 10 years from now, group look backmost and deliberation of the main domiciled OpenAI played during the precocious 2010s/early 2020s arsenic being the org that group disconnected awesome excitement astir and finance successful AGI (and past mislaid its lead to different orgs).”
- Toner successful a connection relayed by Brockman to different OpenAI leaders.
“Because of this shape of lying… arsenic was being reported to me, group successful the institution were copying that behavior, and location was benignant of a civilization of lying and a civilization of, you know, yeah, deceit. And I deliberation for us, arsenic a board… this was conscionable highly concerning.”
- McCauley successful her deposition.
“I thought it would beryllium 1 of the coolest things that humanity could ever build. I was a sci-fi nerd. I publication a batch of books. I watched a batch of sci-fi TV and movies. And, you know, I thought it would beryllium 1 of the astir adjuvant things to thief humanity prosper.”
- Altman, during his deposition, connected why he wanted to subordinate OpenAI.
“I mean, he played a batch of video games.”
- Altman’s consequence to questioning during his deposition astir Musk’s engagement successful the early days of OpenAI.
“I estimate that I walk and person spent each the measurement done astir a 4th of my clip recruiting for OpenAI.”
- Altman during his deposition.
“I deliberation it’s difficult to find group arsenic successful arsenic Elon Musk.”
- Sutskever during his deposition.
“It doesn’t matter who wins if everyone dies.”
- Brockman successful an early speech pinch OpenAI colleagues.
Follow topics and authors from this communicative to spot much for illustration this successful your personalized homepage provender and to person email updates.
![Stanford's Youngest Instructor On Infosec, Ai, Catching Cheaters - Rachel Fernandez [podcast #217]](https://i1.ytimg.com/vi/GmtOxMl39Tc/maxresdefault.jpg "Stanford's Youngest Instructor On Infosec, Ai, Catching Cheaters - Rachel Fernandez [podcast #217]")
English (US) · 
Indonesian (ID) ·