At 4 a.m. on April 10, someone threw a Molotov cocktail at Sam Altman’s San Francisco home. The fire singed an exterior gate. Nobody was hurt. The 20-year-old suspect was arrested an hour later after showing up at OpenAI’s headquarters making arson threats. Four days earlier, The New Yorker had published a sweeping investigation into the OpenAI CEO built on more than 100 interviews and 70 pages of internal memos, Slack messages, and HR documents. The timing was hard to ignore: a firebomb hitting the doorstep of a man whose own colleagues had just called him “unconstrained by truth.”
The Sam Altman New Yorker investigation, authored by Ronan Farrow and Andrew Marantz, is the most detailed account to date of how the CEO of an $852 billion company allegedly built a pattern of deception that stretches back years, across multiple organizations, and through the boardroom that was supposed to keep him in check. For founders reading this from the other side of a cap table, the story is not gossip. It is a case study in what happens when trust erodes at the top of a company and nobody stops it in time.
Last updated: April 2026
What did the New Yorker investigation find about Sam Altman?
The investigation found a documented pattern of dishonesty that spans Altman’s career, not just his time at OpenAI. Farrow and Marantz reviewed previously undisclosed memos assembled by Ilya Sutskever, OpenAI’s former chief scientist, who had compiled roughly 70 pages of annotated Slack messages, internal communications, and HR records before his departure. One memo states plainly: “Sam exhibits a consistent pattern of lying.” A former OpenAI board member told the reporters, “He’s unconstrained by truth.” Y Combinator co-founder Paul Graham was quoted telling colleagues after Altman’s departure from YC: “Sam had been lying to us all the time.”
The allegations are specific. In December 2022, Altman reportedly assured board members that controversial GPT-4 features had been approved by an internal safety panel. Board member Helen Toner later discovered that the most contested features, including one allowing users to fine-tune the model, had never been approved at all. The investigation also details allegations that Altman lobbied against the European Union’s AI Act and California’s state-level AI safety bill while simultaneously testifying before the U.S. Senate in favor of regulation.
For a company now valued at $852 billion and generating an estimated $25 billion in annualized revenue, the gap between public messaging and private behavior described in the report is staggering. This is not a scrappy startup where corners get cut because there is no process. OpenAI has a board, a CFO, a legal team, and safety researchers. The investigation suggests that the process existed but was consistently overridden.
The safety promise OpenAI broke
OpenAI committed 20% of its total computing power to its superalignment team, the group tasked with making sure AI systems smarter than humans could still be controlled. That commitment was public. It was part of how OpenAI distinguished itself from competitors. Insiders told Fortune the real number was closer to 1 to 2%.
In May 2024, both co-leaders of the superalignment team resigned. Sutskever left quietly. Jan Leike, the team’s other co-lead, was blunt on his way out: “Over the past years, safety culture and processes have taken a backseat to shiny products.” He said his team had been “struggling for compute” and that requests for GPU access were repeatedly denied by leadership.
Within weeks, OpenAI disbanded the superalignment team entirely. Some members were reassigned. Others left the company. The 20% compute promise became a footnote.
This is the detail from the investigation that matters most for anyone building a company. A public commitment that is not backed by internal resources is not a strategy. It is marketing. And when the people responsible for holding the company accountable to that commitment leave or are pushed out, the commitment dies with them. Every founder who has ever made a promise to investors, customers, or employees about where resources would go should see this for what it is: a cautionary tale about what happens when your public narrative and your internal operations stop matching.
Was Sam Altman really fired for lying in 2023?
Yes, according to the people who made the decision. On November 17, 2023, OpenAI’s board of directors removed Altman as CEO, stating that he had not been “consistently candid in his communications.” At the time, the public story was chaotic. Employees threatened to quit. Microsoft signaled it would hire Altman directly. Within five days, Altman was reinstated. The board members who fired him were replaced.
The New Yorker investigation reframes that episode entirely. Former board member Helen Toner told the reporters that two OpenAI executives had brought the board screenshots and documentation of what they described as “lying and being manipulative in different situations.” Toner also alleged that executives reported “psychological abuse” from Altman. The 70-page dossier compiled by Sutskever was part of the evidence the board reviewed before making its decision.
The firing was not a personality clash or a philosophical disagreement about the speed of AI development, which is how it was widely reported at the time. Based on the investigation’s account, it was an accountability mechanism triggered by documented evidence of deception. The mechanism worked exactly once. Then it was overridden by a show of financial force from Microsoft, employees, and investors who decided the CEO was more valuable than the governance structure designed to check him.
The lesson here is uncomfortable. Boards exist to protect the company from exactly this scenario. When the board did its job, the market punished it. That sends a clear signal to every startup board in the world: accountability is supported in theory and punished in practice when the stakes are high enough. Good governance habits matter, but they only work if the people around the table are willing to absorb the consequences.
How Altman responded to the investigation
Altman published a blog post on the evening of April 10, hours after the Molotov cocktail attack on his home. He addressed both events. On the investigation, he acknowledged mistakes. “Looking back,” he wrote, “I can identify a lot of things I’m proud of and a bunch of mistakes,” including “being conflict-averse” and handling the 2023 board conflict badly.
He initially described the New Yorker article as “incendiary,” then walked that word back after people on X pointed out the uncomfortable irony, given that someone had just thrown an actual incendiary device at his house. On the broader question of trust, Altman offered no direct rebuttal to the core allegations. He did not dispute the memos. He did not deny the specific incidents. His response focused instead on the future of AI and the need for society to respond to “fear and anxiety about AI.”
From a communications standpoint, this is a textbook redirect. Acknowledge mistakes in general terms. Avoid engaging with specifics. Shift the conversation to a bigger topic where you control the framing. It works well in press cycles. It does not work well in boardrooms, with co-founders, or with investors who need to trust the numbers you put in front of them. Thinking like an entrepreneur means knowing the difference between managing a narrative and managing a relationship.
What should founders take from this?
The founder lesson from the Sam Altman New Yorker investigation is not “don’t lie.” That is obvious. The lesson is that dishonesty compounds, and the compounding happens faster than most people expect.
Altman’s alleged pattern, as described in the investigation, followed a recognizable arc. Small misrepresentations about safety approvals became larger misrepresentations about resource allocation. Those became a culture where, according to Jan Leike, “safety culture and processes have taken a backseat.” The people who objected left. The people who stayed learned that objecting was not rewarded. By the time the board tried to intervene, the company’s financial gravity was too strong for governance to overcome.
Every founder who has stretched a metric in a board deck, downplayed a risk in a fundraising meeting, or told a co-founder that something was further along than it actually was should recognize this arc. The individual acts feel small. The cumulative effect is a company where nobody trusts the information they are given, and the people whose job is to push back stop pushing.
The investigation also surfaces a second, less discussed lesson. Altman’s trajectory at OpenAI shows that charisma and results can sustain a leader through almost any accountability event, up to a point. He survived the board firing. He survived the superalignment team collapse. He survived executive departures. Each time, the explanation was that Altman was too important, the vision too big, the company too valuable to risk disrupting. The New Yorker investigation is a test of whether that narrative can survive documented evidence presented at scale.
For founders at earlier stages, the question is simpler. If you are building a company where you are the only person who knows the full picture, and you are selectively sharing that picture with the people who are supposed to keep you accountable, you are building the same structure Altman built. The only difference is the number of zeros on the valuation.
The AI anxiety problem founders cannot ignore
The Molotov cocktail attack adds a dimension to this story that goes beyond corporate governance. Someone was angry enough about AI to physically attack the home of an AI company CEO. San Francisco police arrested the 20-year-old suspect after he also appeared at OpenAI’s office making threats.
Altman linked the attack to broader AI anxiety in his blog post. That framing is self-serving but not entirely wrong. Public fear about AI is real, measurable, and growing. A February 2026 Pew Research survey found that 52% of Americans say they are more concerned than excited about AI’s role in daily life, up from 38% in 2023. When the CEO of the world’s largest AI company is credibly accused of lying about safety commitments, that fear has somewhere specific to land.
This matters for every founder in AI, and increasingly for founders in any industry where automation is changing the terms of employment. The gap between Altman’s public vision and his company’s internal reality is exactly the kind of story that erodes public trust not just in OpenAI but in the entire sector. Founders who are honest about the limitations and risks of their products are building a moat that Altman apparently chose not to build. Transparency is not a constraint on growth. It is a competitive advantage in a market where trust is collapsing.
How board trust breaks and what it costs
The New Yorker investigation describes a specific mechanism of trust failure that repeats across organizations. First, the CEO controls the flow of information to the board. Second, the board makes decisions based on incomplete or inaccurate information. Third, when someone inside the company tries to correct the record, they face retaliation or isolation. Fourth, the people who could hold the CEO accountable either leave or are removed. Fifth, the CEO’s control becomes self-reinforcing because there is no one left with both the information and the authority to challenge them.
OpenAI followed this pattern precisely. Sutskever compiled the dossier. Toner asked hard questions. The board acted. And then the market, employees, and investors reversed the board’s decision in less than a week. The message was clear: the CEO is the company, and the company is too big to govern.
Building a startup from inception to exit requires trust at every level. The Sam Altman case shows what the endgame looks like when that trust is treated as optional. The company survives. The valuation grows. But the people who were supposed to keep it honest are gone, and the structure that was supposed to prevent exactly this outcome has been dismantled. For any founder reading this, the time to build a culture of honesty with your board is before you are worth $852 billion, not after a reporter with 100 sources comes knocking.
Starting a business in 2026 means building in an environment where trust is the scarcest resource. The founders who treat board transparency as a feature, not a bug, are the ones who will still have boards that function when the pressure arrives.
