Situational Awareness: One Insider's Case for Fast AGI

Leopold Aschenbrenner, who worked on OpenAI's superalignment team, wrote a long essay series in June 2024 arguing that AGI by roughly 2027 is, in his words, strikingly plausible. The argument is an extrapolation of three trend lines rather than a single prediction. Compute for frontier training is growing about half an order of magnitude per year. Algorithmic efficiency is improving at a similar rate. On top of that he adds gains from what he calls unhobbling, the shift from raw chatbot to tool-using agent. Stack those together and he projects a jump comparable to the move from GPT-2 to GPT-4 happening again within a few years, then hundreds of millions of automated researchers compressing a decade of algorithmic progress into well under a year. The essay does not stop at capability. Aschenbrenner argues the buildout implies trillion-dollar compute clusters, warns that frontier labs have almost no real security around model weights and algorithmic secrets, and predicts a nationalized, Manhattan-style effort by the late 2020s. You can read the full series at situational-awareness.ai.

Why it matters

This document shaped a lot of the 2024 to 2026 conversation about AGI timelines, datacenter capex, and AI as a national security problem. Whether or not you accept the timeline, investors, labs, and policymakers now argue using its framing, so it is worth reading the original rather than the summaries.