The Atlantic Profiles the Bitter Rivalry Between Anthropic's Amodei and OpenAI's Altman
Summary
- • The Atlantic feature reveals the deep personal and ideological feud between Amodei and Altman
- • Amodei called Altman a 'two-faced manipulator' and OpenAI 'a shit show'
- • Amodei's team pioneered scaling laws and RLHF before he left to found Anthropic
- • Dispute rooted in OpenAI's shift from nonprofit mission toward commercial enterprise
Details
Amodei joined OpenAI in 2016 as VP of Research
Princeton-educated physicist left Google for OpenAI in 2016; led GPT-2 and GPT-3 development; known for 26-page internal essay 'Big Blob of Compute' outlining his theory of intelligence and AGI
Scaling laws originated with Amodei's team
Kaplan and McCandlish derived equations predicting predictable LLM performance gains from scale (data, compute, parameters); steered OpenAI's research agenda toward large language models
RLHF pioneered by researcher on Amodei's team
Paul Christiano, a researcher under Amodei, developed reinforcement learning from human feedback — the technique that allows LLMs to be steered away from harmful outputs
Amodei viewed Altman as 'two-faced manipulator'
Also called OpenAI co-founder Greg Brockman 'power-hungry and reckless'; described OpenAI as 'a shit show run by low-integrity people' to a person close to him at the time
OpenAI's commercial pivot drove Amodei out
Amodei opposed converting OpenAI from a nonprofit to a commercial enterprise; believed it strayed from the mission to benefit humanity unconstrained by financial return
Anthropic founded on safety-first ideological split
Amodei left OpenAI and co-founded Anthropic, centering the lab on safety and mission alignment over commercialization — the direct product of his break with Altman
Feud continues to shape AI industry's core divide
The Amodei-Altman rivalry drives competing approaches to safety standards, deployment speed, and AI governance at the two most influential frontier labs
The Atlantic feature on the Amodei-Altman feud and its origins, September 2026
What This Means
The Atlantic's deep-dive feature exposes the personal and philosophical rupture that produced the two dominant AI labs of the 2020s. Amodei's foundational role in developing scaling laws and RLHF — the two techniques that underpin modern LLMs — gives the feud particular historical weight: these aren't just commercial rivals, but ideological opponents with competing visions of what responsible AI development requires. This public accounting of the split matters because the Anthropic-OpenAI rivalry continues to shape safety norms, regulatory debates, and the pace of AI deployment industry-wide.