🛡️

Dario Amodei

Original nameDario Amodei

American AI Researcher, CEO of Anthropic

Those Who Turned Papers Into What a Billion People Use · Industry & Product Builders

AnthropicThe "Constitutional AI" MethodThe Industrialization of AI Safety

Who they are

Dario Amodei (1983– ) is an American AI researcher, former VP of research at OpenAI, who directed the development of early large language models such as GPT-2 and GPT-3. In 2021 he, with his sister Daniela Amodei and several other researchers also from OpenAI, co-founded Anthropic, the core motive being that the capability scaling of large language models had already outpaced the industry’s research on their safety and controllability, and that a team treating safety issues more cautiously should take part in driving the technology’s development. The Claude series of large language models Anthropic thereafter developed uses its own "Constitutional AI" training method, letting the model self-critique and correct by a preset set of principles to reduce reliance on large-scale human-annotated feedback. Amodei has long publicly written on the loss-of-control risks AI may bring, while also advocating cautiously but firmly continuing to push the frontier of technical capability, and this "cautious yet competitive" positioning has made Anthropic a double competitor of OpenAI, Google DeepMind, and others in both business and ideas.

Primary sourcesAnthropic’s official founding statement (2021)Amodei’s public essay "Machines of Loving Grace"

Key stories

Leaving OpenAI to Build "the Safer One"

Amodei and several colleagues including his sister Daniela, holding that OpenAI, while pushing model-capability scaling, had failed to keep pace in its attention to and investment in safety issues, chose to resign collectively and found Anthropic, the company name itself taken from "human-centered." This split is thereafter seen as a landmark event of the fundamental tension in the AI industry between "how fast to push technical capability" and "how cautiously to treat safety risk," and Anthropic thereafter became an important rival in direct business and ideological competition with OpenAI.

Relationships

Ally

🧯Jan Leike　Leike, after resigning as head of OpenAI’s "superalignment" team in 2024, joined Anthropic to continue AI-safety research

Nemesis

💬Sam Altman

Echoes today

Below are how modern works borrow or reinterpret this name or story — not the original material. The two differ, so keep them apart.

The industry position of "both safe and competitive"While stressing safety and caution, Anthropic has also kept releasing frontier models with industry competitiveness, a position often seen as an important reference for examining the industry’s core dispute of "whether safety and commercialization can both be had."

Appears in

The Generative-AI Explosion and the Industry ContestThe Split of Anthropic and "Constitutional AI"

← Back to Figures