General · EdgepediaSeptember 17, 202612 min read

Dario Amodei

Dario Amodei (born 1983) is an American AI researcher and entrepreneur, co-founder and CEO of Anthropic, who previously served as vice president of research at OpenAI, where he helped pioneer reinforcement learning from human feedback and led the GPT-3 project.¹ He is described as perhaps the purest believer in the scaling laws among AI research leaders.¹

Key fact	Detail
Education	PhD in computational neuroscience, Princeton University; dissertation won the Hertz Thesis Prize²^•¹
OpenAI career	Joined 2016; VP of Research; led GPT-3, which consumed 50–60% of the company's compute¹
Anthropic founding	Left OpenAI in December 2020 with his sister Daniela and other colleagues; company founded 2021³^•⁴
Signature research	Co-author and guide of the 2020 "Scaling Laws for Neural Language Models" paper⁵
Company scale	Anthropic raised $30 billion at a $380 billion post-money valuation in February 2026, with $14 billion in run-rate revenue⁶
Safety framework	The Responsible Scaling Policy ties model capability thresholds (ASL levels) to required safeguards, with the board ultimately accountable to the Long Term Benefit Trust⁷
Estimated wealth	Forbes estimated the seven billionaire cofounders, led by Dario and Daniela Amodei, at $7 billion apiece after the Series G⁸

Early life and education

Amodei was born in San Francisco in 1983 to a Jewish mother and an Italian father, and as a young man was interested almost entirely in mathematics and physics.¹ His father Riccardo died in 2006, and the loss prompted Amodei to shift his graduate studies at Princeton from theoretical physics to biology, with the aim of addressing human illness.¹

At Princeton he worked in Professor Michael Berry's retina lab, where he co-invented a new sensor for retinal signals, and his dissertation won the Hertz Thesis Prize, an award for work with real-world applications.¹ He completed a PhD in computational neuroscience.²

Research career and scaling laws

After about ten months at Google, Amodei joined OpenAI in 2016 and worked on AI safety, helping to pioneer reinforcement learning from human feedback (RLHF), the technique used to fine-tune GPT-2.¹ As OpenAI's vice president of research he led the GPT-3 project.²^•¹ The GPT-3 project received 50–60% of the company's entire compute and represented a 100x scale-up costing tens of millions of dollars.¹

The scaling hypothesis. The hypothesis was formalized in the 2020 paper "Scaling Laws for Neural Language Models," a collaboration between Johns Hopkins University and OpenAI led by Jared Kaplan and Sam McCandlish, with Amodei as a co-author who provided guidance throughout the project.⁵ The paper found that language model loss scales as a power law with model size, dataset size, and training compute, with trends spanning more than seven orders of magnitude, and concluded that larger models are more sample-efficient, so compute-optimal training uses very large models stopped before convergence.⁵ Amodei has said that he and his Anthropic cofounders were among the first to document and track these scaling laws.⁹

Founding of Anthropic

Accounts differ on the size of the departing group: one profile describes Amodei, Jack Clark, Daniela Amodei, Chris Olah, and a handful of others leaving in December 2020,¹ while another reports the Amodeis quitting along with about 15 likeminded colleagues.¹⁰

Within a few weeks Amodei had collected seven cofounders: himself, Daniela, Jack Clark, Sam McCandlish, Tom Brown, Chris Olah, and Jared Kaplan, with Ben Mann added later as an eighth.³ Anthropic was founded in 2021 and has focused on enterprise sales, while OpenAI has been largely consumer-focused through ChatGPT.⁴

Daniela Amodei, Dario's sister, is a cofounder and, with him, leads the company; Forbes lists the sibling pair first among Anthropic's seven billionaire cofounders.⁸ Forbes estimates each of the seven cofounders owns just over 1.8% of the company.⁸

Constitutional AI and safety research

Constitutional AI, described in Anthropic's 2022 paper, trains a harmless assistant through self-improvement in which the only human oversight is a list of rules or principles, rather than human labels identifying harmful outputs.¹¹ The method has two phases: a supervised phase of self-critique and revision fine-tuning, and a reinforcement learning phase using RL from AI Feedback (RLAIF), in which a preference model is trained on AI-generated preferences rather than human ones.¹¹ The result is an assistant that is harmless but non-evasive, engaging with harmful queries by explaining its objections, and the approach reduces the need for human labels.¹¹

Amodei describes Constitutional AI as one of Anthropic's core innovations, aspects of which have since been adopted by other AI companies, in which post-training is guided by a central document of values and principles.⁹ In its reinforcement-learning phase, it uses AI-generated preference judgments based on written principles rather than human preference labels.¹¹

Leadership and business trajectory

Anthropic's funding has grown rapidly. A compiled funding history records a $3.5 billion Series E in March 2025 led by Lightspeed at a $61.5 billion valuation, and a $13 billion Series F in September 2025 led by ICONIQ at $183 billion.¹² In February 2026 the company raised $30 billion in Series G funding led by GIC and Coatue at a $380 billion post-money valuation, with D. E. Shaw Ventures, Dragoneer, Founders Fund, ICONIQ, and MGX co-leading or participating.⁶ A subsequent $65 billion Series H in May 2026, led by Altimeter, Dragoneer, Greenoaks, and Sequoia at a $965 billion valuation, is reported by the same compiled source.¹² An earlier profile reported that Amodei had brought in nearly $20 billion of investment, including $8 billion from Amazon and $3 billion from Google.¹

On the product side, Claude is available on all three of the world's largest cloud platforms, AWS Bedrock, Google Cloud Vertex AI, and Microsoft Azure Foundry, and runs on AWS Trainium, Google TPUs, and NVIDIA GPUs.⁶ Claude Code, made generally available in May 2025, reached a run-rate revenue of over $2.5 billion, more than doubling since the beginning of 2026.⁶ Amodei's compensation as CEO was US$18 million in the year covered by the company's IPO filing, which Reuters placed in the middle ranks of tech CEOs.¹³

By the numbers

Run-rate revenue: $14 billion as of the February 2026 Series G announcement, growing over 10x annually in each of the previous three years.⁶
Claude Code: over $2.5 billion run-rate revenue; one analysis estimated 4% of all GitHub public commits worldwide were authored by Claude Code, double the share from one month prior.⁶
Valuation progression: $61.5 billion (March 2025), $183 billion (September 2025), $380 billion (February 2026), $965 billion (May 2026, per compiled disclosures).¹²
2025 financials per IPO filing: about $4.6 billion in revenue, $7.33 billion spent on compute, an $8 billion operating loss, and a $42 billion net loss that included a $34 billion item.¹⁴
Cofounder stakes: just over 1.8% each, worth an estimated $7 billion apiece after the Series G.⁸

Two revenue figures conflict and are reported here as such. The company said in February 2026 that revenue the prior year reached roughly $10 billion,⁴ while IPO-filing figures reported by Reuters put 2025 revenue at about $4.6 billion.¹⁴

Safety advocacy and public positions

The Responsible Scaling Policy. Anthropic's policy uses an AI Safety Level (ASL) system loosely modeled after the internationally recognized biosafety level (BSL) system, with each level carrying an if-then structure tying dangerous capabilities to required safeguards.⁷ ASL-3 is triggered when models become operationally useful for catastrophic misuse in CBRN (chemical, biological, radiological, nuclear) areas, requiring strong security so non-state actors cannot steal model weights and ensuring deployed models never produce CBRN-risk information even under expert red-teaming.⁷ ASL-4 will be triggered when AI systems approach near-human autonomy or become the main source in the world of at least one serious global security threat, such as bioweapons.⁷ The policy is a formal directive of Anthropic's board, which is ultimately accountable to a Long Term Benefit Trust, an external panel of experts with no financial stake in the company.⁷ Amodei has described the framework as an engineering and governance regime in which capability evaluations are run at each capability jump rather than on an annual or quarterly schedule.² As of mid-2025, Anthropic's measurements indicated that LLMs may provide substantial uplift relevant to bioweapons acquisition, perhaps doubling or tripling the likelihood of success, which led the company to release Claude Opus 4 and subsequent Opus and Sonnet models under ASL-3 protections.⁹

Policy engagement. Amodei testified before the US Senate Judiciary Committee on July 26, 2023, discussing safety research and the launch of Claude 2 for US users.¹⁵ He has supported state transparency legislation, citing California's SB 53 and New York's RAISE Act as bills Anthropic backed that passed.⁹ In a letter warning that AI safety measures are out of step with AI's rapid progress, he argued the government "should have the power to block or deter deployment of the model if it is determined, in light of third-party assessment, to present unacceptable risks," and he announced a $200 million investment in research on AI's impact on society.¹⁶ In July 2026 he said he does not oppose open-weight models but fears Chinese AI, and supports creating a model safety testing organization, ideally with global participation including China; he called safety testing for the most capable models regardless of country of origin "close to a consensus" and noted the Trump administration had moved in that direction.¹⁷

Essays and public writing

In 2024 Amodei published "Machines of Loving Grace," a cautiously optimistic essay about "powerful AI" that analogizes its deployment to "a country of geniuses in a datacenter" and predicts, with a great deal of effort, the defeat of most diseases.¹⁸ Written partly in response to perceptions that Amodei was an AI "doomer," the essay sketches outcomes if "everything goes right."¹⁸ Its techno-optimism drew criticism from safety advocates: one critic characterized his follow-up essay, "The Adolescence of Technology," as "a reluctant political manifesto" whose "shoulds and maybes" sit in awkward contrast with an account of AI progress full of "wills and cans."¹⁸ In that follow-up, Amodei argued that fully addressing AI risks requires not just investing in risk prevention but pacing the rate of capabilities advancement so risk prevention has time to keep up.¹⁹

How it compares with other AI leaders

Amodei is described as perhaps the purest believer in scaling laws among AI research leaders, while peers such as Google DeepMind CEO Demis Hassabis and Meta's then chief AI scientist Yann LeCun suggest the industry needs further breakthroughs.¹ Hassabis was deeply skeptical of the scaling laws that Sam Altman and Amodei rode to power.²⁰ On the accelerationist side, OpenAI investors Marc Andreessen and Gary Tan identify as AI accelerationists, Altman has signaled sympathy for that worldview, and LeCun has expressed broadly accelerationist views, in contrast with Amodei's safety-focused stance.¹⁰ In a CBS News interview Amodei said AI development has been on an exponential curve that is starting to get steeper, calling it "a warning sign" while saying it does not mean panic or shutting AI down, and that "for too long the industry lied" about AI risks.²¹ DER SPIEGEL frames the Amodei-Altman relationship as a rivalry between OpenAI's former head of research, once one of Altman's most important engineers, and the chief rival company, in a race whose competitive dynamics increase AI risks.²²

Reception and criticism

Several episodes have been cited as tension between Anthropic's safety commitments and its commercial behavior. Earlier this year, Anthropic clashed with the Pentagon because Amodei demanded the Defense Department refrain from deploying Anthropic's technology for widespread domestic surveillance of Americans or in autonomous weaponry.²³ February 2025 saw Anthropic officially drop its commitment to halt training of ever more capable models when their abilities exceeded what the firm could comprehend and steer, a shift that recast the policy from a binding rule into a mere aspiration.¹⁰ A leaked internal memo from Amodei to staff took a tone sharply different from "Machines of Loving Grace," excoriating OpenAI and describing the reported provisions of its deal as "safety theater."²⁴ Five years after its founding, Anthropic, with thousands of employees, the Pentagon feud, and a trillion-dollar IPO on the horizon, barely resembles the company its founders set out to create, in the assessment of one long-form history.³

Amodei's response has been to frame the strategy as deliberate. He argues that enterprise customers' low tolerance for AI incidents means commercial incentives point toward caution, since a single high-profile failure in a regulated industry can set back enterprise AI adoption by years, and he describes Anthropic's continued frontier work as "a calculated bet, and I'm honest that it's a bet."² In his "pace the frontier" essay he frames the goal as making safety something AI companies compete on, "a race to the top," while acknowledging accusations of hype, "doomerism," and regulatory capture, and says Anthropic has tried to prioritize caution over speed and prudence over profit.¹⁹ The Atlantic assesses his slowdown stance as likely genuine, noting the company was founded by researchers who left OpenAI and Google over disagreements about AI development paths.²³

References

Topic: Encyclopedia › Technology and the built world › Computing and digital systems › Modern AI: foundation models, generative AI, and the AI industry › AI companies, people, and products › AI founders and executives › Applied AI startup founders

Initially written Sep 17, 2026 · Reviewed: Oct 15, 2026 · Edited: Oct 15, 2026 · Last review: Oct 15, 2026

Notice something wrong?