Who is Leopold Aschenbrenner?

A researcher who was an early member of OpenAI's Superalignment team. In *Situational Awareness* (2024), a series of essays, Aschenbrenner argued that AGI could arrive around 2027 and that the US and China are in a race for superintelligence, which the free world must win.

Within a few years AI smarter than humans will appear, and nations will race each other for it. This person's series of essays drew a huge response by laying out that outlook concretely, with numbers and a timeline. This article introduces Aschenbrenner as someone who discusses the future of AI from a position different from that of the people on the side of the book Plurality.

In short

Leopold Aschenbrenner is a researcher who was one of the first members of OpenAI's "Superalignment" team, which researched the safety of superintelligence, when it began [1].

In June 2024, Aschenbrenner published a series of essays, Situational Awareness: The Decade Ahead. They argue that AGI (artificial general intelligence, AI that can do a broad range of work as humans do) could appear around 2027, and that because superintelligence would give a decisive economic and military advantage, the free world must win the race [2]. → Superintelligence

On the Dwarkesh Podcast in the same month, June 2024, Aschenbrenner said he was launching an investment firm that would bet on AGI and superintelligence, with anchor investments from Patrick Collison, John Collison, Daniel Gross and Nat Friedman [1]. What the firm has done since then is not covered in this site's materials.

Connection to Plurality

No direct mention by the book, Weyl or Tang. In the book Plurality (in its various language editions) and in the records of Glen Weyl's and Audrey Tang's talks and conversations, no passage has been found that discusses Aschenbrenner or Situational Awareness by name. The one exception is the conversation described below, in which the host mentioned the essays by name.

Buterin's mention. In "d/acc: one year later" (January 2025), Vitalik Buterin wrote that regulation of AI hardware is being considered, as with export controls, in a spirit of trusting one's own side but not the other side. As an example, he cites Aschenbrenner's argument that the US should first race to gain a decisive advantage, and then effectively make China accept an agreement limiting the number of computers it can run. Buterin sees this approach as dangerous, liable to combine the drawbacks of both racing and centralization. If limits are needed, in Buterin's view, the effort should go into limiting everyone equally and building mechanisms for cooperation, rather than one side dominating the others [3]. → d/acc

In a question to Tang. On the program "Doom Debates" in January 2026, the host, Liron Shapira, mentioned recent writings such as Situational Awareness, and another paper whose name he had forgotten that said nations will really go after each other's AI power, and noted that Taiwan will be central to the future economy [4]. Asked next about international treaties and an AI "pause button", Audrey Tang replied that, worried about AI's power to strongly persuade people, Tang had signed the "Red Lines" declaration (a call to set lines that must not be crossed). Tang went on to say that, as in Taiwan's anti-fraud measures, society can practice drawing small, close-to-home red lines, not censoring speech but making visible who is speaking and whether there is a digital signature, and that such practice becomes the foundation for large international treaties in the future [4]. This was not a remark about Aschenbrenner personally.

In the debate over the future of AI. Situational Awareness drew a number of responses, including Max Tegmark's "The Hopium Wars: the AGI Entente Delusion" [5]. This site also covers it, as one of the representative documents of a position different from Plurality's, in the article on the AI futures debate.

Main ideas

From here on, this section draws on Aschenbrenner's own writings and statements.

Situational Awareness (2024). The introduction writes that the AGI race has already begun. The outlook is that by 2025–26 AI will outpace many college graduates, and that by the end of the decade there will be superintelligence in the true sense [2].

The chapter summaries (by the author) are as follows [2].

Chapter I: counting the growth of computation and algorithmic efficiency in orders of magnitude ("OOMs"), it argues that AGI by 2027 is strikingly plausible.
Chapter II: it describes an "intelligence explosion" in which hundreds of millions of AGIs automate AI research itself, compressing a decade of algorithmic progress into less than a year.
Chapter IIId, "The Free World Must Prevail": it argues that superintelligence will give a decisive economic and military advantage, and that China is not yet out of the race.
Chapter IV, "The Project": it predicts that as the AGI race intensifies, the national security establishment will get involved, and that by 2027–28 the US government will embark on some form of AGI project.

If you want to win the race, solve alignment (2023). Alignment is research on making AI act in line with human intentions and values. Aschenbrenner wrote that if we want to beat China in the AGI race, we need a better alignment plan. The reasoning is that without a solution good enough for broad agreement that superhuman AGI is safe, we may well not be able to proceed [6]. In another post the same day, Aschenbrenner warned that while there are plausibly about 100,000 researchers in the world working to advance AI capabilities, there are only about 300 alignment researchers [7].

Growth and existential risk (2020). In a working paper for the Global Priorities Institute at the University of Oxford, Aschenbrenner argued, using an economic model of the direction of technology, that existential risk may follow a hump shape (an inverted U). We may now be in a "time of perils", having technologies that could destroy humanity but not yet being rich enough to spend enough on safety. Speeding up growth during this period, the paper argues, first raises risk but raises the chances of long-run survival (only the abstract was retrieved) [8].

Burkean longtermism (2021). A piece headed by a passage from the eighteenth-century thinker Edmund Burke's Reflections on the Revolution in France, to the effect that society is a partnership between the living, the dead and those yet to be born. Aschenbrenner named "Burkean longtermism" a view close to, but different from, that of friends who are "longtermists" deeply concerned with the distant future, and listed as its propositions concern for the distant future, gratitude for what we have inherited, and humility about the limits of what we can know, among others [9].

How Aschenbrenner left OpenAI (Aschenbrenner's own account). On the Dwarkesh Podcast, Aschenbrenner said that OpenAI had told employees it fired him for leaking information. By his account, the "leak" was showing a brainstorming memo on the safety and security measures needed on the way to AGI to three outside researchers. He also said that at the time of the firing, a memo on security was stated explicitly to be a major reason. This is Aschenbrenner's side of the story [1].

What is debated

Buterin's view. As seen above, Buterin regards as dangerous the path of first going for a decisive advantage and then making the other side accept limits. His position is that equal limits and efforts at cooperation should come first [3].
Tegmark and others' responses. Situational Awareness drew a number of responses on LessWrong and elsewhere. In an October 2024 piece, Tegmark took up the strategy of democracies banding together to gain an advantage over China in AI (the "entente" strategy), naming Anthropic CEO Dario Amodei and Aschenbrenner among its proponents. Tegmark argues that, whatever one's ethical or geopolitical preferences, the strategy is fundamentally flawed and contrary to US national security interests: if the US and China race for AGI, the only winner will be the machines [5].

Sources

Dwarkesh Patel, "Leopold Aschenbrenner — 2027 AGI, China/US Super-Intelligence Race, & The Return of History" (2024-06-04) — https://www.dwarkesh.com/p/leopold-aschenbrenner
Leopold Aschenbrenner, Situational Awareness: The Decade Ahead, introduction and contents (2024-06, excerpt) — https://situational-awareness.ai/
Vitalik Buterin, "d/acc: one year later" (2025-01-05) — https://vitalik.eth.limo/general/2025/01/05/dacc2.html
Audrey Tang and Liron Shapira, "Doom Debate" (2026-01-20) — https://archive.tw/2026-01-20-doom-debate ［CC0］
Max Tegmark, "The Hopium Wars: the AGI Entente Delusion" (LessWrong, 2024-10-13) — https://www.lesswrong.com/posts/oJQnRDbgSS8i6DwNu/the-hopium-wars-the-agi-entente-delusion
Leopold Aschenbrenner, "Want to win the AGI race? Solve alignment." (2023-03-29) — https://www.forourposterity.com/want-to-win-the-agi-race-solve-alignment/
Leopold Aschenbrenner, "Nobody's on the ball on AGI alignment" (2023-03-29) — https://www.forourposterity.com/nobodys-on-the-ball-on-agi-alignment/
Leopold Aschenbrenner, "Existential Risk and Growth" (GPI Working Paper No. 6-2020, abstract) — https://web.archive.org/web/20250617201207/https://globalprioritiesinstitute.org/leopold-aschenbrenner-existential-risk-and-growth/
Leopold Aschenbrenner, "Burkean Longtermism" (2021-07-26) — https://www.forourposterity.com/burkean-longtermism/

We welcome error reports and questions ("What is…?"). Submissions become public issues.

Report an error in this article ・ Ask a question