Dario Amodei

Who they are

Dario Amodei is CEO and co-founder of Anthropic — a Princeton-trained biophysicist who left OpenAI's VP of Research role in 2021 to found the company, and wrote 'Machines of Loving Grace,' a landmark essay on AI's positive potential.

Person

Dario started at Caltech before transferring to Stanford, where he finished a BS in Physics in 2006, then went to Princeton for a PhD in Biophysics, completed in 2011 — the kind of training that makes you think about complex systems at a mechanistic level. He did a postdoc at Stanford School of Medicine, then passed through Applied Minds and Schlumberger (where he worked as a geophysicist and consultant) before landing at Google Brain as a Senior Research Scientist. He moved to Baidu in 2014–2015, working under Andrew Ng on Deep Speech, then joined OpenAI in 2016, climbing from Research Scientist to Team Lead for AI Safety to VP of Research — the person running the research engine at the moment GPT-3 was being built. In 2021 he left, taking a cohort of colleagues with him including his sister Daniela, and co-founded Anthropic from scratch at seed stage. He writes at darioamodei.com — his most-cited piece is 'Machines of Loving Grace' (October 2024), a long-form essay mapping AI's potential impact on biology, neuroscience, governance, and work; he also published 'The Urgency of Interpretability.' He testified before the US Senate Judiciary Committee's Subcommittee on Privacy, Technology, and the Law in 2023 on AI regulation. The through-line: physicist who kept moving closer to the hardest, least-solved problem in the field — and then decided to own it.

Company

Anthropic closed a $65 billion Series H in May 2026 at a $965 billion post-money valuation, surpassing OpenAI (valued at $852 billion as of March 2026) to become the world's most valuable private AI company. That capped a run of three large rounds in under a year — Series F ($13 billion at $183 billion, September 2025), Series G ($30 billion at $380 billion, February 2026), then Series H. The company reported a $47 billion annualized run rate as of May 2026, with Claude Code alone reaching $2.5 billion in annualized revenue by February 2026. Anthropic confidentially submitted a draft S-1 to the SEC around June 2026, signaling a potential IPO in fall 2026, advised by Wilson Sonsini. The period has not been without friction: the US government ordered suspension of its most powerful models (Fable 5 and Mythos 5) for foreign nationals over cybersecurity concerns, and Anthropic is engaged in a legal dispute with the Pentagon over the application of its safety safeguards to lethal autonomous weapons use. In July 2026, Anthropic launched Ode with Anthropic — a joint venture with Blackstone, Hellman & Friedman, and Goldman Sachs — raising $1.5 billion in committed capital.

Market

Anthropic operates in the frontier AI model market against OpenAI, Google DeepMind, Meta, Mistral, and Cohere. In enterprise large language models specifically, Anthropic claimed a 40% market share in 2025 (surpassing OpenAI's 27% and Google's 20%) and leads in coding AI with a 42% market share. The broader competitive dynamic is increasingly geopolitical: US export controls, a bifurcating US-China AI ecosystem, and regulatory scrutiny of circular financing arrangements with cloud partners (AWS, Google, Nvidia) are structural forces shaping how Anthropic can grow internationally.

Network

Dario's closest working partner is his sister Daniela Amodei, President and co-founder of Anthropic, who handles operations, partnerships, and external functions — they co-departed OpenAI together. The founding team also includes Chris Olah (interpretability research), Jared Kaplan (scaling laws), Jack Clark (policy), Sam McCandlish, and Ben Mann. Earlier in his career, Dario worked directly under Andrew Ng at Baidu on Deep Speech.

Daniela Amodei· President and Co-founder, Anthropic
Chris Olah· Co-founder, Anthropic (interpretability research)
Jared Kaplan· Co-founder, Anthropic
Jack Clark· Co-founder, Anthropic (policy)
Andrew Ng· Former manager at Baidu; AI researcher
Ben Mann· Co-founder, Anthropic
Sam McCandlish· Co-founder, Anthropic

How they likely show up

PhD in Biophysics followed by a geophysics stint at Schlumberger → comfort moving between very different technical domains; likely approaches AI problems with a first-principles, mechanistic mindset rather than pattern-matching from precedent.
Left OpenAI as VP of Research — the top research seat — to start over at seed stage → high conviction, willing to take large personal risk when he disagrees with the institutional direction; not someone who stays for the title.
Writes long-form essays at darioamodei.com on AI safety and interpretability → thinks publicly, in structured arguments; expects conversations to be substantive and conceptually precise.
Testified before the US Senate Judiciary Committee and is active in the Frontier Model Forum → comfortable in adversarial, high-stakes institutional settings; treats policy engagement as a core part of the job, not a distraction.
Founded Anthropic as a Delaware Public Benefit Corporation → the legal structure signals that safety and mission are load-bearing commitments, not marketing; he will be sensitive to anything that looks like safety theater.
Role type is founder-operator, not hired executive → he sets the agenda; he is not optimizing for a boss's approval, he is optimizing for the research and the mission as he defines them.

Conversation tips

→ Reference a specific argument from 'Machines of Loving Grace' or 'The Urgency of Interpretability' — he will immediately know whether you read it or are bluffing, and it opens a real conversation rather than a pitch.
→ Ask about interpretability as a technical program, not just a talking point — Chris Olah's work is foundational to Anthropic's identity and Dario cares about the underlying science.
→ The Pentagon dispute and the Mythos/Fable 5 suspension are live and sensitive — if you raise them, do so with specificity and a genuine question, not as a gotcha; he has clearly thought harder about these tradeoffs than most people in the room.
→ Don't use the IPO as an opener — it is clearly in motion but he has not gone on record treating it as the story; lead with product or research instead.
→ He moved from physics to biophysics to AI safety — he responds to people who can make rigorous cross-domain connections, not people who speak only in AI-industry shorthand.

Toolbox

Openers

Open on 'The Urgency of Interpretability' — he published it at darioamodei.com as a companion to 'Machines of Loving Grace,' and it signals that interpretability is not just a research bet but an existential priority; asking what progress actually looks like in practice gets straight to what he cares about.
Reference the Mythos model situation — Anthropic withheld the release of Mythos over cybersecurity concerns in May 2026, triggering a geopolitical incident and calls from US leadership for AI company cooperation; it is the sharpest recent example of the tension between safety commitments and commercial and government pressure.
Bring up Claude Code hitting $2.5 billion in annualized revenue by February 2026 — it is the most concrete proof point that the safety-first positioning is compatible with enterprise scale, and it's a product milestone he almost certainly has views on.

Discovery questions

You've argued for stronger government oversight of AI while simultaneously fighting the Pentagon over safety safeguards in weapons applications — how do you draw the line between the oversight you want and the oversight you're resisting?
Claude Code reached $2.5 billion in annualized revenue faster than almost any enterprise product — did that change how you think about the relationship between the research agenda and the product roadmap?
In 'Machines of Loving Grace' you lay out a fairly specific positive vision for AI's impact on biology and neuroscience — how much of that vision is driving what Anthropic actually builds versus what sits in the essay as an aspiration?

Avoid

Don't open with generic AI safety framing or platitudes about 'responsible AI' — he has spent years making precise technical and philosophical arguments about alignment and interpretability, and vague safety-speak will read as either uninformed or insincere.

Make it yours