Writings
The pieces that define the thinking. Essays, papers, and public remarks from both Dario and Daniela Amodei.
Essays by Dario
Machines of Loving Grace 2024
A detailed vision of how AI could transform health, science, economics, and governance — if developed responsibly. Not a prediction, but a conditional optimism.
Read essayThe Adolescence of Technology 2025
On the awkward middle period of powerful technologies — too capable to ignore, too immature to trust fully. A framework for navigating the current moment.
Read essayOn DeepSeek and Export Controls 2025
A clear-eyed analysis of compute geopolitics and why export controls matter for the trajectory of AI development.
Read essayDaniela on leadership & safety culture
Anthropic's Responsible Scaling Policy 2023
Co-authored framework defining capability thresholds and the safeguards required at each level. Daniela's operational perspective shaped its real-world enforceability.
Read on AnthropicBuilding Trust in AI 2024
Daniela's public remarks on why trust with governments, enterprises, and the public requires operational transparency — not just technical capability.
Core Views on AI SafetyScaling a Safety-First Company 2025
On maintaining founding principles during hypergrowth — from 30 people to 1,000+ while keeping safety culture intact. Covers hiring, decision-making structures, and institutional design.
Anthropic ResearchResearch papers
Constitutional AI 2022
Introduced RLAIF — training AI systems using written principles rather than relying solely on human feedback. Foundational to Claude's alignment approach.
Read on arXivScaling Laws for Neural Language Models 2020
Demonstrated that model performance follows predictable power laws with respect to compute, data, and parameters. Shaped how the industry thinks about scaling.
Read on arXivSleeper Agents 2024
Research showing that deceptive behaviors can persist through standard safety training. Important for understanding the limits of current alignment techniques.
Read on arXivToward Monosemanticity 2023
Interpretability research identifying individual features in neural networks. A step toward actually understanding what models compute.
Read articleOp-eds
AI Needs Basic Transparency Rules 2025
Published in the New York Times. Argues for minimum disclosure standards in AI development.
NYTTrump Can Keep America's AI Advantage 2025
Published in the Wall Street Journal. On maintaining U.S. leadership in AI through strategic investment and policy.
WSJ