Sam Altman & OpenAI: The Organisation That Changed Everything

“AI is the most consequential technology that humanity has ever had the opportunity to develop. … OpenAI’s mission is to ensure that artificial general intelligence — AI systems that are generally smarter than humans — benefits all of humanity.”

San Francisco, California. November 17, 2023. A Friday afternoon. Sam Altman is about to be fired.

He has been the Chief Executive Officer of OpenAI since 2019, having joined the organisation in 2015 as President. Under his leadership, OpenAI has gone from a small research laboratory with about a hundred employees and an ambitious mission statement to the most consequential AI organisation in the world — the creator of GPT-4, the deployer of ChatGPT, a company valued at tens of billions of dollars that has put conversational AI in the hands of hundreds of millions of people.

The board of directors — a small group that includes the organisation’s chief scientist Ilya Sutskever, the researcher Helen Toner, the AI safety advocate Tasha McCauley, and the entrepreneur Adam D’Angelo — has made the decision in private. Altman is informed via a video call that the board has lost confidence in his leadership, that he has not been consistently forthright with them, and that his position is terminated, effective immediately.

The announcement, released publicly a few hours later, says: “Mr. Altman’s departure follows a deliberative review process by the board, which concluded that he was not consistently candid in his communications with the board, hindering its ability to exercise its responsibilities.”

What follows is one of the most dramatic five days in the history of technology. By the following Wednesday, Sam Altman is back.

The board crisis of November 2023 is both the most dramatic external manifestation of OpenAI’s internal tensions and a window into the deeper questions that the organisation’s existence raises: What does it mean to build transformative AI? Who should control it? What do we owe to the future when we are building technology that might be the most consequential in human history?

This is the story of OpenAI. It is also, inescapably, the story of Sam Altman.

---

Chicago to Stanford: The Formation of a Builder

Samuel Harris Altman was born on April 22, 1985, in Chicago, Illinois. His parents were a dermatologist and a homemaker. He grew up in St. Louis, Missouri, in a household that was intellectually engaged and moderately comfortable. He came out as gay at the age of seventeen — a formative experience in a time and place where this was not without social cost, and one that he has discussed as shaping his relationship to authenticity and to the courage to say unpopular things.

He arrived at Stanford in 2003 to study computer science and lasted two years before dropping out to start a company. The company was Loopt — a location-sharing social networking application that was ahead of its time in 2005, when smartphones were not yet ubiquitous and the social infrastructure for location sharing did not exist. Loopt was funded by Y Combinator in 2005, in the second batch of companies the incubator accepted. It was one of Paul Graham’s early experiments in a format that would transform startup culture.

Loopt never became the dominant social networking application that Altman envisioned. The timing was wrong, the market was not ready, and the specific form of social interaction that location sharing enabled proved less compelling than other forms. Loopt was eventually acquired by Green Dot Corporation in 2012 for approximately $43 million — a modest outcome for a venture-backed company that had been operating for seven years.

But the Loopt experience gave Altman a specific kind of knowledge that subsequent success would be built on: the knowledge of what it was like to build a company, to recruit people, to manage investors, to make decisions under uncertainty, to fail in specific ways and to recover. It also gave him a connection to Paul Graham and Y Combinator that would prove consequential.

In 2011, Altman joined Y Combinator as a part-time partner. In 2014, he became President of Y Combinator — the managing partner responsible for the day-to-day operation of the most influential startup accelerator in Silicon Valley. He was twenty-eight years old.

---

Y Combinator and the Education of a President

Altman’s years at Y Combinator — from 2011 to 2019, when he left to focus full-time on OpenAI — were the formative period of his professional identity. The specific things he learned at YC, the specific people he met, and the specific perspective on technology and its relationship to society that YC embodied all shaped how he approached OpenAI and how he understood its mission.

Y Combinator under Altman became something different from what it had been under Paul Graham. Graham’s YC was small, opinionated, and intensely focused on a specific theory of startups — the theory that the best companies were started by technical founders solving real problems, funded early and quickly with small amounts of money, and allowed to grow rapidly by finding product-market fit. Graham articulated this theory in a series of essays that became the foundational text of Silicon Valley startup culture.

Under Altman, YC expanded dramatically — the batch sizes grew, the network grew, the institutional ambitions grew. YC began to think about its role not just as a startup accelerator but as an organisation that could affect the trajectory of technology and, through technology, the trajectory of society. This grander ambition — technology as a means of civilisational improvement — was something that Altman brought from his own temperament and that he would carry into OpenAI.

The people Altman met through YC, and the ideas he encountered in the Silicon Valley ecosystem, exposed him to the specific intellectual tradition of effective altruism and to the broader set of concerns about existential risk from advanced AI that were being developed by researchers at institutions like MIRI (Machine Intelligence Research Institute) and the Future of Humanity Institute. These concerns would shape his thinking about what AI was and what it required — and would eventually lead to the founding of OpenAI.

---

The Founding of OpenAI: A Bet Against Concentration

OpenAI was founded in December 2015. The founding team included Sam Altman, Elon Musk, Greg Brockman, Ilya Sutskever, Wojciech Zaremba, John Schulman, and several others. The initial funding commitment was $1 billion, though the actual amount invested was considerably less in the early years.

The founding motivation was specific and, in retrospect, revealing. Several of the founders believed that AI was approaching a level of capability that would make it potentially transformative and potentially dangerous. They believed that the development of this technology was inevitable — that if they did not develop it, others would. And they believed that it was better to have safety-focused researchers at the frontier of AI capability than to leave that frontier entirely to organisations whose primary motivation was commercial profit.

The specific concern about concentration was important. In 2015, Google had acquired DeepMind and had a substantial head start in AI research. Facebook had established FAIR (Facebook AI Research) with Yann LeCun at its head. The worry was that if the most powerful AI systems were developed entirely within large technology companies, the control of these systems — and the direction of their development — would be determined by the commercial interests of those companies.

OpenAI’s solution was to structure itself as a nonprofit — specifically as a 501(c)(3) charitable organisation — with the explicit mission of “ensuring that artificial general intelligence benefits all of humanity.” The nonprofit structure was supposed to provide an accountability mechanism that commercial organisations lacked: the board of directors would be obligated to the mission, not to shareholders, and decisions about how AI was developed and deployed would be evaluated against the mission rather than against revenue.

This was a genuinely unusual institutional form for a technology research organisation, and the founders took the unusual structure seriously. The articles of incorporation described the mission in strong terms: OpenAI would pursue AI development in a way that was safe, and in a way that ensured the benefits were broadly distributed rather than concentrated. The board was given specific powers to enforce these commitments.

---

The Early Years: Research Without Revenue

OpenAI’s early years were years of research without revenue, funded primarily by the initial donor commitments and by a series of subsequent fundraising efforts. The research was serious and productive — OpenAI hired some of the best machine learning researchers in the world and produced important results in reinforcement learning, language modelling, and AI safety.

The reinforcement learning work was particularly notable. OpenAI Gym, published in 2016, was a toolkit for training and evaluating reinforcement learning agents on a standardised set of environments — a contribution that became standard infrastructure for RL research throughout the field. OpenAI Five, the reinforcement learning system that played the video game Dota 2 at championship level by 2018, was a demonstration of the power of large-scale reinforcement learning applied to complex, real-world games.

The language modelling work produced the GPT series — GPT-1 in 2018, GPT-2 in 2019. These were the systems that demonstrated the power of large-scale Transformer pre-training, and they established OpenAI as a frontier AI research organisation.

But the research was expensive. Training GPT-2 at the scale OpenAI envisioned required substantial computing resources — thousands of GPU-hours, significant infrastructure costs. Training GPT-3 at 175 billion parameters would cost tens of millions of dollars in computing alone. The $1 billion commitment from the founders was not sufficient for the kind of research OpenAI was pursuing.

This financial reality drove the most consequential institutional change in OpenAI’s history: the 2019 transition from a pure nonprofit to a “capped profit” company structure.

---

The Capped Profit Pivot: When Mission Met Money

In 2019, OpenAI restructured. The nonprofit remained as the controlling entity — the OpenAI Nonprofit — but a new for-profit subsidiary, OpenAI LP, was created to raise capital from investors. The for-profit structure allowed OpenAI to offer equity to employees and to raise money from investors who expected financial returns.

The “capped profit” element of the structure was designed to maintain some connection to the nonprofit mission. Returns for investors and employees were capped — limited to a specific multiple of their investment, after which excess returns would flow back to the nonprofit. The idea was that OpenAI could raise the capital it needed while limiting the extent to which profit motives would distort its mission.

Microsoft invested $1 billion in OpenAI LP in 2019, cementing a partnership that would prove enormously consequential. Microsoft provided computing resources through its Azure cloud platform, which allowed OpenAI to train the large models that its research required. In exchange, Microsoft received a share of OpenAI’s commercial revenues and, eventually, the right to deploy OpenAI’s models in its own products.

The capped profit structure satisfied the letter of OpenAI’s mission commitments more than it satisfied the spirit. The reality of the new structure was that OpenAI was now an organisation with investors who expected returns, employees who had equity compensation, and a commercial partner (Microsoft) whose interests were tied to OpenAI’s commercial success. These stakeholders created pressures toward commercial success that were structurally similar to the pressures that a conventional for-profit technology company faced.

Sam Altman was the person most responsible for navigating these pressures, and his approach reflected his specific temperament and his specific theory of how to pursue OpenAI’s mission in the new institutional environment.

---

Sam Altman’s Theory of OpenAI: Speed as Safety

Altman’s theory of how to pursue OpenAI’s mission in the commercially constrained environment was, essentially, that being at the frontier was necessary for having influence over how AI developed. If OpenAI fell behind — if Google or Anthropic or some other organisation became the leader in AI capability — OpenAI would lose its ability to shape the field’s direction. Being commercially successful was necessary for generating the resources to remain at the frontier. And being at the frontier was necessary for having the influence that the mission required.

Altman’s genuine belief — and it appears to be genuine, not merely strategic — was that AGI was coming regardless of what OpenAI did, and that the question was not whether to develop it but how to develop it in a way that was most likely to be beneficial. In this belief, he was in a specific tradition that went back to the founding: the founders of OpenAI had believed that the development of powerful AI was inevitable, and that the right response was to be at the frontier rather than to cede it to less safety-conscious actors.

Whether this belief was correct — whether OpenAI’s presence at the frontier made the development of powerful AI safer or whether it accelerated development in ways that increased risk — is one of the central contested empirical questions in AI.

---

GPT-4 and the Commercial Breakthrough

The release of GPT-4 in March 2023 — and its deployment through ChatGPT — represented the commercial breakthrough that Altman had been working toward. GPT-4 was substantially more capable than its predecessors: more accurate, more consistent, better at reasoning, less likely to hallucinate, and more capable at following complex instructions. It was the most capable language model publicly deployed at the time of its release.

The combination of GPT-4’s capability and ChatGPT’s accessibility — the simple chat interface that made the model available to anyone without technical expertise — produced the most rapid adoption of any consumer technology product in history. A hundred million users in two months. A cultural moment in which AI shifted from a specialised technical topic to a mainstream public concern.

For OpenAI, the commercial success was both validating and complicated. Validating because it demonstrated that the technology OpenAI had been developing was genuinely valuable — that real users found real benefit from it. Complicated because the commercial success accelerated the commercial pressures that the capped profit structure had been designed to manage.

The commercial success also created competitive responses that accelerated the pace of AI development across the industry. Google, Microsoft, Meta, and dozens of startups all accelerated their AI development in response to ChatGPT’s success. The competitive dynamics that OpenAI’s commercial success triggered produced exactly the kind of racing dynamic that OpenAI’s founders had hoped to prevent.

Altman responded to these competitive dynamics by accelerating OpenAI’s own development. The logic was the same: if others were going to develop powerful AI anyway, it was better to be ahead of them. But the logic was also a bit circular: OpenAI’s commercial success had helped trigger the race, and OpenAI’s response to the race was to race harder.

---

Ilya Sutskever and the Safety Dimension

One of the most important relationships in OpenAI’s history was the partnership between Altman and Ilya Sutskever — OpenAI’s Chief Scientist, the person most responsible for the technical direction of the organisation’s AI research.

Sutskever was one of the most technically gifted AI researchers of his generation — a student of Hinton’s who had co-developed AlexNet, who had been recruited to Google after the acquisition of DNNresearch, and who had left Google to co-found OpenAI with Altman, Musk, and the others. His technical judgment was widely respected, and he had been the primary architect of the research programmes that produced the GPT series and, through them, ChatGPT and GPT-4.

Sutskever had also, through his years at OpenAI, developed increasingly serious concerns about AI safety — about whether the systems OpenAI was building were genuinely aligned with human values, and about whether the pace of development was outrunning the ability to ensure that the systems were safe. These concerns were genuine and deeply held, not strategic positioning.

The tension between Altman’s “move fast to remain at the frontier” philosophy and Sutskever’s deepening safety concerns was one of the most important internal tensions at OpenAI through 2022 and 2023. It was also one of the least visible externally — both men understood the value of presenting a unified public face, and the disagreements were managed internally rather than aired publicly.

Until November 2023.

---

The Board Crisis: Five Days That Changed OpenAI

The events of November 17-22, 2023 — the five days between Altman’s firing and his reinstatement — are the most consequential episode in OpenAI’s history and one of the most dramatic in the history of the technology industry.

The board’s decision to fire Altman was made by a majority vote that included Ilya Sutskever. The specific reasons given — that Altman had not been consistently candid with the board — were vague, and the board did not release more specific information about what Altman had done or not done that had led to the loss of confidence.

In the absence of specific information, speculation was extensive. Some believed the firing was primarily about safety concerns — that the board, representing the nonprofit mission, had concluded that Altman’s commercial orientation was undermining OpenAI’s commitment to safe AI development. Some believed it was about specific incidents of miscommunication or concealment. Some believed it was about the board’s concerns about Altman’s ambitions to raise enormous amounts of capital for AI investment in ways that would transform OpenAI’s mission and governance.

The immediate aftermath was chaos. Altman and Greg Brockman, OpenAI’s President, were initially expected to go to Microsoft, which was prepared to create a new AI research unit for them. Microsoft’s stock rose on the announcement of this plan. OpenAI employees signed an open letter demanding the board’s resignation and Altman’s reinstatement, and the letter was signed by approximately 90% of the workforce — a remarkable expression of employee sentiment that effectively threatened the organisation’s ability to function without Altman.

Sutskever quickly reversed his position, signing the employee letter and expressing regret for his role in the firing decision. This reversal was both a sign of how rapidly the situation had evolved and of the complexity of Sutskever’s own position — he had apparently concluded that the consequences of the firing were worse for OpenAI’s mission than whatever had motivated the decision.

By November 22, Altman was reinstated as CEO. The board that had fired him resigned and was replaced by a new board with different composition. Altman had survived the crisis, and OpenAI continued with him in charge.

---

What the Crisis Revealed

The November 2023 board crisis revealed several things about OpenAI and about the broader challenge of governing transformative AI organisations.

The governance structure was inadequate. The specific governance arrangement that OpenAI had designed — nonprofit board controlling a for-profit subsidiary — proved unable to manage the tension between mission and commercial success at the scale OpenAI had reached. The board that fired Altman was small, lacked the institutional authority to enforce its decision in the face of employee opposition and investor pressure, and ultimately could not execute its decision effectively.

The safety-capability tension was real and unresolved. Whatever specifically motivated the board’s decision, the crisis was an expression of the underlying tension between the safety concerns that some board members had and the commercial and capability orientation that Altman represented. This tension was not resolved by Altman’s reinstatement — it was merely deferred.

Commercial success created leverage. Altman’s reinstatement was possible largely because the employees, investors, and Microsoft all had interests aligned with keeping Altman. The commercial success that OpenAI had achieved under Altman’s leadership had created a set of stakeholders who would act to protect that success. The nonprofit mission, which the board was supposed to protect, had less leverage than the commercial interests when they came into conflict.

---

Altman’s Public Persona: The Advocate and the Executive

One of the most distinctive features of Altman’s leadership of OpenAI is the specific public persona he has cultivated — the combination of AI advocate, existential risk acknowledger, and commercial executive that he presents in interviews, speeches, and congressional testimony.

Altman has been unusually candid, by the standards of technology company executives, about the potential risks of the technology he is building. He has testified before Congress that AI could be “the most transformative and potentially dangerous technology ever developed by humanity.” He has said publicly that there is a real probability that AI development goes badly for humanity. He has engaged seriously with the concerns of AI safety researchers in ways that most commercial AI company leaders have not.

This candour is unusual and its interpretation is contested. His admirers see it as evidence of genuine intellectual honesty — a willingness to acknowledge the risks of the technology he is building rather than simply selling its benefits. His critics see it as strategic — a way of co-opting the AI safety conversation and making OpenAI seem safety-conscious while actually pursuing commercial objectives that may increase risk.

---

The Race to AGI: What OpenAI Believes It Is Building

OpenAI’s mission statement is to ensure that artificial general intelligence — AGI — benefits all of humanity. AGI, as OpenAI defines it, is “highly autonomous AI systems that outperform humans at most economically valuable work.”

The question of whether AGI, so defined, is a coherent concept — whether there is a single threshold that AI systems cross to become “generally intelligent” — is contested. Many AI researchers believe that intelligence is not a single capability that can be achieved at a single threshold, but a collection of capabilities that AI systems will achieve at different rates in different domains. On this view, “AGI” is a misleading framing that obscures the specific capabilities that specific systems have and the specific capabilities they lack.

Altman and OpenAI take AGI seriously as a concept and as a goal. They believe that AI systems are approaching a level of general capability that will be qualitatively different from the domain-specific AI that currently exists. They believe this development is coming in years, not decades. And they believe that how this development is managed — who builds the systems, with what safety precautions, with what governance — will be one of the most consequential decisions in human history.

This belief — genuine, held by serious people at OpenAI — shapes the organisation’s strategic decisions in specific ways. The urgency to remain at the frontier is partly a product of this belief: if AGI is coming soon, being at the frontier during the transition matters enormously. The investment in safety research is partly a product of this belief: if AGI is coming soon, figuring out how to make it safe is urgent.

It also shapes the critics’ concerns. If AGI is genuinely coming soon, the competitive dynamics triggered by OpenAI’s commercial success — the racing between organisations to develop more powerful models — are exactly the kind of dynamics that make safe development of AGI most difficult. Racing reduces the time available for careful safety research. Racing creates pressure to deploy before safety is fully understood. Racing concentrates capability development in a small number of organisations with enormous resources and enormous competitive incentives.

---

The Microsoft Partnership: Commercial Entanglement

The Microsoft partnership that began with the 2019 investment and deepened with subsequent investments — Microsoft eventually committed approximately $13 billion to OpenAI — transformed the organisation’s financial position and its competitive capabilities while also creating entanglements that critics found concerning.

The partnership gave OpenAI access to Microsoft’s Azure cloud infrastructure at a scale that no academic research group could have matched. The ability to train GPT-4 and its successors depended on access to thousands of A100 and H100 GPUs over months of training — computing that would have cost hundreds of millions of dollars at market rates and that Microsoft provided through the partnership.

In exchange, Microsoft received the right to deploy OpenAI’s models in its products and to offer them through Azure. Bing Chat, Copilot, and the AI features in Microsoft’s Office and developer tools were all powered by OpenAI’s models. The commercial deployment generated revenue for OpenAI and visibility for Microsoft.

The entanglement raised governance questions. Microsoft was OpenAI’s primary commercial partner, its largest investor, and its primary computing infrastructure provider. The extent to which this entanglement influenced OpenAI’s decisions — about which models to develop, what safety constraints to impose, how fast to deploy — was not transparent. Critics worried that Microsoft’s commercial interests would influence OpenAI in ways that were not compatible with the safety-focused mission.

The November 2023 board crisis amplified these concerns. Microsoft’s readiness to hire Altman and Brockman after their firing — and its clear preference for Altman’s leadership over the board’s decision — was an expression of the commercial alignment between Microsoft’s interests and Altman’s strategy. The board, which represented the nonprofit mission, had less leverage than the investor and computing infrastructure partner.

---

The Critics: What the Safety Community Says

The relationship between OpenAI and the AI safety research community — the researchers who have been working on technical approaches to AI alignment and the advocates who have been arguing for more cautious AI development — has been complex and often tense.

OpenAI’s safety credentials come from several sources: its founding mission, its investment in safety research through teams like OpenAI Safety and Alignment teams, and Altman’s public statements about the risks of the technology. These credentials are real: OpenAI has done important work on RLHF, on interpretability, on the development of safety evals, and on the broader technical programme of AI alignment.

But critics within the safety community argue that the commercial orientation of OpenAI under Altman has compromised these credentials. The release of GPT-4 and ChatGPT — systems that were deployed to hundreds of millions of users before the safety questions surrounding them were fully understood — is cited as evidence of the commercial pressures overriding safety concerns. The competitive dynamics that OpenAI’s commercial success has triggered are cited as evidence that OpenAI’s strategy of moving fast to remain at the frontier has backfired.

The founding of Anthropic — by Dario Amodei, Daniela Amodei, and several other former OpenAI researchers — was explicitly motivated by disagreements about safety at OpenAI. The Anthropic founders believed that OpenAI was not taking safety seriously enough, that commercial pressures were pushing the organisation to deploy systems before they were adequately understood, and that a different approach — one that prioritised safety research more explicitly and moved more cautiously on deployment — was necessary.

The relationship between OpenAI and Anthropic is now publicly civil but privately competitive. They compete for talent, for customers, for computing resources, and for influence in the AI governance conversation. Both claim to be the safety-focused option relative to Google and Meta. Whether either is as safety-focused as it claims is an empirical question that the public cannot fully evaluate from outside the organisations.

---

The Future: What Altman Wants to Build

What Altman appears to want to build — based on his public statements, his strategic decisions, and the trajectory of OpenAI — is a company that occupies the most powerful position in the AI industry at the moment when AI becomes transformative, and uses that position to ensure that the transformation is broadly beneficial.

The plan has several components. Technical leadership: remain at the frontier of AI capability, training the most powerful models, deploying them most widely. Commercial success: generate sufficient revenue to sustain the research that technical leadership requires, and to attract and retain the talent that the research requires. Governance and policy influence: use OpenAI’s position as a frontier AI organisation to influence how AI is governed — what standards are required, what regulation is appropriate, how international competition in AI is managed.

The critics of this plan argue that the first two components — technical leadership and commercial success — are in tension with the third. The competitive dynamics that commercial success creates, and that technical leadership requires, are exactly the dynamics that make thoughtful governance most difficult. An organisation that is racing to be first with the most powerful model is not well-positioned to advocate for the kind of measured, collaborative, internationally coordinated approach to AI governance that the risks might require.

---

The Honest Assessment: What OpenAI Has Done

OpenAI’s record, evaluated honestly, is mixed in ways that reflect the genuine complexity of what the organisation is trying to do.

What OpenAI has done well. OpenAI has produced the most capable publicly available AI systems in the world, and has deployed them in ways that have created genuine value for hundreds of millions of users. It has invested substantially in safety research, producing important technical contributions to RLHF, to interpretability, to safety evaluation, and to the broader technical programme of AI alignment. It has engaged constructively with AI governance conversations, providing technical expertise to government discussions and advocating for regulatory frameworks that, while imperfect, are better than the alternatives being advocated by some.

What OpenAI has done less well. OpenAI’s governance structure, designed to maintain accountability to its mission, proved inadequate when tested by the November 2023 board crisis. The commercial pressures that the organisation’s success created have demonstrably influenced its decisions in ways that are not always transparently compatible with the safety mission. The competitive dynamics that its commercial success has triggered may have accelerated AI development in ways that have increased rather than decreased risk.

What remains uncertain. Whether OpenAI’s approach — being at the frontier while investing in safety — is net positive or net negative for the safety of AI development is genuinely uncertain. Reasonable people with access to the same information disagree. The answer will depend partly on how AI development unfolds — whether the risks that safety researchers worry about materialise, whether the safety research that OpenAI and others have invested in proves adequate to address those risks, and whether the governance frameworks that are established in the next few years are adequate to manage the technology.

---

Sam Altman as a Historical Figure

How history will judge Sam Altman depends on what AI development produces — on whether the trajectory that he has helped accelerate leads to broadly beneficial outcomes or to significant harms.

If AI development goes well — if the most powerful AI systems are broadly beneficial, if the safety research that has been invested in proves adequate, if the governance frameworks that are established are effective — Altman will be remembered as one of the most consequential builders in the history of technology. The person who made ChatGPT, who made large language models accessible to hundreds of millions of people, who navigated the most difficult institutional challenges of the AI era with sufficient skill to keep the most important AI organisation functioning and at the frontier.

If AI development goes badly — if the racing dynamics contribute to deployment of systems that are not adequately safe, if the concentration of AI capability in a small number of organisations produces governance failures, if the harms of AI significantly outweigh its benefits — Altman will be remembered differently. As the person who accelerated the development of technology whose risks he acknowledged but whose deployment he prioritised over safety.

These are not equivalent outcomes, and Altman’s responsibility for which outcome materialises is real and significant. He is the person most responsible for the specific pace and character of AI deployment through the most consequential period of AI’s development. That responsibility is not diminished by the argument that if he had not done it, someone else would have.

Whether he has discharged this responsibility well — whether the specific decisions he has made have been, on balance, the decisions most likely to lead to beneficial outcomes — is a judgment that the available evidence does not unambiguously support or refute. It is a judgment that history will eventually render.

---

The Organisation That Changed Everything

Whatever judgment history eventually renders on Altman personally, OpenAI’s place in the history of AI is assured. It is the organisation that deployed ChatGPT and thereby made conversational AI a mainstream reality. It is the organisation whose GPT series established the scaling paradigm that has defined frontier AI research. It is the organisation whose commercial success triggered the competitive response that has transformed the AI industry.

These are genuine and consequential contributions. The world that OpenAI has helped make — in which hundreds of millions of people interact daily with AI systems capable of genuine language understanding — is a world that would not exist, or would exist later and differently, without OpenAI’s specific contributions.

Whether that world is better or worse than the one that preceded it is a question that cannot be answered yet. The technology is too new, the consequences too diffuse, the counterfactual too uncertain. What can be said is that the world is different — genuinely, consequentially different — and that OpenAI is among the primary authors of that difference.

Sam Altman is the person who led OpenAI through the period in which that difference was made. He is consequential in the specific sense that his decisions mattered — that different decisions would have produced different outcomes. In a field and an era where many decisions matter enormously, his have mattered among the most.

---

Further Reading

---

---