Dario Amodei wants outside testers inside frontier AI companies

His own site says he is CEO of Anthropic, “a public benefit corporation dedicated to building AI systems that are steerable, interpretable and safe”.

Dario Amodei is the CEO of Anthropic, a public benefit corporation dedicated to building AI systems that are steerable, interpretable and safe.

His site says he was a postdoctoral scholar at the Stanford University School of Medicine.

Dario earned his doctorate degree in biophysics from Princeton University as a Hertz Fellow, and was a postdoctoral scholar at the Stanford University School of Medicine.

His site says that before joining OpenAI he worked at Google Brain as a Senior Research Scientist.

Before joining OpenAI, he worked at Google Brain as a Senior Research Scientist.

His site says he was Vice President of Research at OpenAI, where he led the development of GPT-2 and GPT-3.

Previously, Dario served as Vice President of Research at OpenAI, where he led the development of large language models like GPT-2 and GPT-3.

darioamodei.com is his own site: Anthropic’s statement in his name of 21 October 2025 links to his essay “Machines of Loving Grace” there, and the site says he is Anthropic’s CEO.

Dario Amodei is the CEO of Anthropic, a public benefit corporation dedicated to building AI systems that are steerable, interpretable and safe.

His site calls him “the co-inventor of reinforcement learning from human feedback” and lists two essays and four short posts.

He is also the co-inventor of reinforcement learning from human feedback. ... Essays: The Adolescence of Technology; Machines of Loving Grace. Short posts: We Must Pace the Frontier; Policy on the AI Exponential; The Urgency of Interpretability; On DeepSeek and Export Controls

anthropic.com/company lists him on Anthropic’s board of directors.

Anthropic Board of Directors Dario Amodei, Daniela Amodei, Yasmin Razavi, Reed Hastings, Chris Liddell, and Vas Narasimhan.

The page links to Anthropic’s own accounts on YouTube, LinkedIn and X.

https://www.youtube.com/@anthropic-ai https://www.linkedin.com/company/anthropicresearch https://x.com/AnthropicAI

The Hertz Foundation lists him as a 2007 Hertz Fellow and as co-founder and chief executive officer of Anthropic.

2007 HERTZ FELLOW Co-Founder and CEO, Anthropic ... Dario Amodei is co-founder and chief executive officer at Anthropic, an artificial intelligence (AI) safety and research company

The Hertz Foundation lists his undergraduate studies at Stanford University.

EDUCATION Graduate Studies Princeton University Biophysics, Physics ... Undergraduate Studies Stanford University

PhD, Princeton University, 2011; the record names the thesis “Network-Scale Electrophysiology: Measuring and Understanding the Collective Behavior of Neural Circuits” and the advisers Michael J. Berry and William Bialek, and files it under the Department of Physics.

He is an author of Deep Speech 2, from Baidu Research’s Silicon Valley AI Lab (authors listed alphabetically).

OpenAI announced on 29 December 2020 that he was leaving after nearly five years, and credited him with collaborating with the team to build GPT-2 and GPT-3 and with co-leading the direction of its research with Ilya Sutskever.

The Series A announcement ($124 million) says the company is led by Dario Amodei (CEO) and Daniela Amodei (President).

He is the first of six authors of “Concrete Problems in AI Safety”, where his affiliation is given as Google Brain.

Authors:Dario Amodei, Chris Olah, Jacob Steinhardt, Paul Christiano, John Schulman, Dan Mané ... (PDF, page 1:) Dario Amodei Chris Olah Jacob Steinhardt Google Brain Google Brain Stanford University Paul Christiano UC Berkeley John Schulman OpenAI Dan Mané Google Brain

The paper lists “avoiding reward hacking” as one of five practical research problems on accident risk, with “avoiding side effects”.

We present a list of five practical research problems related to accident risk, categorized according to whether the problem originates from having the wrong objective function ("avoiding side effects" and "avoiding reward hacking"), an objective function that is too expensive to evaluate frequently ("scalable supervision"), or undesirable behavior during the learning process ("safe exploration" and "distributional shift").

With Danny Hernandez he wrote that since 2012 the compute used in the largest AI training runs had been increasing exponentially with a 3.4-month doubling time.

He is a co-author of the GPT-2 paper, “Language Models are Unsupervised Multitask Learners”.

The Scaling Laws paper says he “provided guidance throughout the project”.

Contributions: Jared Kaplan and Sam McCandlish led the research. [...] Dario Amodei provided guidance throughout the project.

Scaling Laws: the abstract says a language model’s loss scales as a power law with model size, dataset size and the compute used for training.

The loss scales as a power-law with model size, dataset size, and the amount of compute used for training, with some trends spanning more than seven orders of magnitude.

The GPT-3 paper says: “Dario Amodei designed and led the research.”

Dario Amodei designed and led the research.

GPT-3: the abstract describes a language model with 175 billion parameters, tested on tasks given only a few examples as text.

Specifically, we train GPT-3, an autoregressive language model with 175 billion parameters, 10x more than any previous non-sparse language model, and test its performance in the few-shot setting. For all tasks, GPT-3 is applied without any gradient updates or fine-tuning, with tasks and few-shot demonstrations specified purely via text interaction with the model.

He is a co-author of “Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback” (April 2022).

[Submitted on 12 Apr 2022] Title:Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback Authors:Yuntao Bai, Andy Jones, ... Catherine Olsson, Dario Amodei, Tom Brown, Jack Clark, Sam McCandlish, Chris Olah, Ben Mann, Jared Kaplan

HH-RLHF: the abstract says the paper applies preference modelling and RLHF to fine-tune language models to act as helpful and harmless assistants.

Abstract:We apply preference modeling and reinforcement learning from human feedback (RLHF) to finetune language models to act as helpful and harmless assistants.

He is one of 51 authors of “Constitutional AI: Harmlessness from AI Feedback” (December 2022), published under CC BY 4.0.

[Submitted on 15 Dec 2022] Authors:Yuntao Bai, Saurav Kadavath, ... Ben Mann, Dario Amodei, Nicholas Joseph, Sam McCandlish, Tom Brown, Jared Kaplan

Constitutional AI: the abstract describes training a harmless assistant with no human labels for harmful outputs, the only human oversight being a list of principles; a supervised phase of self-critique and revision; and an RL phase in which AI comparisons of pairs of answers train a preference model used as the reward.

We experiment with methods for training a harmless AI assistant through self-improvement, without any human labels identifying harmful outputs. The only human oversight is provided through a list of rules or principles, and so we refer to the method as 'Constitutional AI'. The process involves both a supervised learning and a reinforcement learning phase. In the supervised phase we sample from an initial model, then generate self-critiques and revisions, and then finetune the original model on revised responses. In the RL phase, we sample from the finetuned model, use a model to evaluate which of the two samples is better, and then train a preference model from this dataset of AI preferences. We then train with RL using the preference model as the reward signal

In “Machines of Loving Grace” he names five areas: biology and physical health; neuroscience and mental health; economic development and poverty; peace and governance; work and meaning.

In “On DeepSeek and Export Controls” he writes that DeepSeek’s releases “make export control policies even more existentially important”.

In “The Urgency of Interpretability” (April 2025) he gives Anthropic’s goal of getting to “interpretability can reliably detect most model problems” by 2027.

Anthropic is doubling down on interpretability, and we have a goal of getting to “interpretability can reliably detect most model problems” by 2027.

He writes that not being able to see inside the models is “literally a legal blocker” for some uses, such as mortgage assessments.

In fact, for some applications, the fact that we can’t see inside the models is literally a legal blocker to their adoption—for example in mortgage assessments where decisions are legally required to be explainable.

In “The Urgency of Interpretability” (April 2025) he wrote: “literally a legal blocker”

In fact, for some applications, the fact that we can’t see inside the models is literally a legal blocker to their adoption—for example in mortgage assessments where decisions are legally required to be explainable.

In “The Urgency of Interpretability” (April 2025) he wrote: “When a generative AI system does something, like summarize a financial document, we have no idea, at a specific or precise level, why it makes the choices it does—why it chooses certain words over others, or why it occasionally makes a mistake despite usually being accurate.”

Generative AI is not like that at all. When a generative AI system does something, like summarize a financial document, we have no idea, at a specific or precise level, why it makes the choices it does—why it chooses certain words over others, or why it occasionally makes a mistake despite usually being accurate. As my friend and co-founder Chris Olah is fond of saying, generative AI systems are grown more than they are built

“The Adolescence of Technology” (January 2026) is subtitled “Confronting and Overcoming the Risks of Powerful AI”.

The Adolescence of Technology Confronting and Overcoming the Risks of Powerful AI January 2026

He writes that enterprises often choose between “cost savings” and “innovation”.

Enterprises often have a choice between “cost savings” (doing the same thing with fewer people) and “innovation” (doing more with the same number of people).

Section 4 of “The Adolescence of Technology” is headed “Player piano”, and in it he names two problems he is worried about: labor market displacement and concentration of economic power.

4. Player piano ... Labor market disruption ... There are two specific problems I am worried about: labor market displacement, and concentration of economic power.

In “The Adolescence of Technology” (January 2026) he wrote: “Enterprises often have a choice between “cost savings” (doing the same thing with fewer people) and “innovation” (doing more with the same number of people).”

Second, AI companies have a choice in how they work with enterprises. The very inefficiency of traditional enterprises means that their rollout of AI can be very path dependent, and there is some room to choose a better path. Enterprises often have a choice between “cost savings” (doing the same thing with fewer people) and “innovation” (doing more with the same number of people).

In “The Adolescence of Technology” (January 2026) he wrote: “being creative about ways to reassign employees within companies may be a promising way to stave off the need for layoffs”

Third, companies should think about how to take care of their employees. In the short term, being creative about ways to reassign employees within companies may be a promising way to stave off the need for layoffs.

In “The Adolescence of Technology” (January 2026) he wrote: “This is a topic that I warned about very publicly in 2025, where I predicted that AI could displace half of all entry-level white collar jobs in the next 1–5 years, even as it accelerates economic growth and scientific progress.”

There are two specific problems I am worried about: labor market displacement, and concentration of economic power. Let’s start with the first one. This is a topic that I warned about very publicly in 2025, where I predicted that AI could displace half of all entry-level white collar jobs in the next 1–5 years, even as it accelerates economic growth and scientific progress.

In “The Adolescence of Technology” (January 2026) he wrote: “some didn’t see the 1–5-year time range and thought I was claiming AI is displacing jobs right now (which I agree it is likely not)”

Many CEOs, technologists, and economists agreed with me, but others assumed I was falling prey to a “lump of labor” fallacy and didn’t know how labor markets worked, and some didn’t see the 1–5-year time range and thought I was claiming AI is displacing jobs right now (which I agree it is likely not).

With “Policy on the AI Exponential” (June 2026), Anthropic released a legislative proposal on frontier model testing and a policy framework for job displacement, for which he writes Anthropic intends to provide “substantial financial backing”.

Along with this essay, Anthropic is releasing a legislative proposal on frontier model testing and a policy framework for job displacement, for which we intend to provide substantial financial backing.

He writes that Anthropic works with customers on new use cases and new sources of revenue so they can do more with their existing workforce, rather than focusing solely on cost savings.

As a company, Anthropic always does as much as it can to work with customers to find creative new use cases and new sources of revenue that allow them to do more with their existing workforce, rather than focusing solely on cost savings (which often means reducing the workforce).

In “Policy on the AI Exponential” (June 2026) he wrote: “However, now the risks are clearly here. It is time to go beyond transparency to more serious and binding regulation of AI.”

However, now the risks are clearly here. It is time to go beyond transparency to more serious and binding regulation of AI.

In “We Must Pace the Frontier” (September 2026) he proposes that each frontier AI company give embedded third-party evaluators “ongoing, employee-like access”, and writes that Anthropic is committing to this step now on its own; for Anthropic’s own external review team, he writes that reviewers should have the right to publish key findings “without editorial control by Anthropic”.

Each frontier AI company commits to giving ongoing, employee-like access to a team of embedded third-party evaluators (such as METR), whose role is to verify adherence to safety practices and commitments, report incidents, and help assess the alignment of not just completed AI models but training pipelines and processes. [...] Anthropic is unilaterally committing to this step now. [...] External reviewers should have the right to publish key findings about risk levels, incidents, practices, and the access they received or didn’t receive — without editorial control by Anthropic.

In “We Must Pace the Frontier” he describes the embedded evaluators’ role: to verify adherence to safety practices and commitments, report incidents, and help assess the alignment of models and training processes.

whose role is to verify adherence to safety practices and commitments, report incidents, and help assess the alignment of not just completed AI models but training pipelines and processes.

In “We Must Pace the Frontier” (September 2026) he wrote: “To be clear, pacing does not mean halting model training or technical progress, but ensuring companies take adequate time to align and safeguard their models, and for third party evaluators to confirm this.”

To be clear, pacing does not mean halting model training or technical progress, but ensuring companies take adequate time to align and safeguard their models, and for third party evaluators to confirm this.

In “We Must Pace the Frontier” (September 2026) he wrote: “We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain.”

We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain.

Anthropic launched Claude more broadly on 14 March 2023, through a chat interface and an API.

Anthropic published its Responsible Scaling Policy on 19 September 2023, calling it “a series of technical and organizational protocols”.

In prepared remarks dated 1 November 2023 he wrote that he spent 10-20% of his time on the RSP for three months, wrote “multiple drafts from scratch” and devised and proposed the ASL system.

Nov 1, 2023 ... First, deep executive involvement is critical. As CEO, I personally spent 10-20% of my time on the RSP for 3 months—I wrote multiple drafts from scratch, in addition to devising and proposing the ASL system.

In the same remarks he called responsible scaling policies a prototype for regulation, not a substitute for it.

Finally, I’d like to discuss the relationship between RSPs and regulation. RSPs are not intended as a substitute for regulation, but rather a prototype for it.

In “Dario Amodei’s prepared remarks from the AI Safety Summit on Anthropic’s Responsible Scaling Policy” (1 November 2023) he wrote: “RSPs are not intended as a substitute for regulation, but rather a prototype for it.”

Dario Amodei’s prepared remarks from the AI Safety Summit on Anthropic’s Responsible Scaling Policy. Nov 1, 2023 ... Finally, I’d like to discuss the relationship between RSPs and regulation. RSPs are not intended as a substitute for regulation, but rather a prototype for it.

On 25 July 2023 he testified to the Senate Judiciary Subcommittee on Privacy, Technology, and the Law, at “Oversight of A.I.: Principles for Regulation”; his opening statement calls for safety tests, including by third parties, before release.

OVERSIGHT OF A.I.: PRINCIPLES FOR REGULATION ... HEARING BEFORE THE SUBCOMMITTEE ON PRIVACY, TECHNOLOGY, AND THE LAW of the COMMITTEE ON THE JUDICIARY UNITED STATES SENATE ... JULY 25, 2023 ... STATEMENT OF DARIO AMODEI, CHIEF EXECUTIVE OFFICER, ANTHROPIC ... Second, we recommend a testing and auditing regime for new and more powerful models. ... New AI models should have to pass a rigorous battery of safety tests before they can be released to the public at all, including tests by third parties and national security experts in Government.

In his opening statement at “Oversight of A.I.: Principles for Regulation” (25 July 2023), as printed in the official hearing record S. Hrg. 118-108, he said: “New AI models should have to pass a rigorous battery of safety tests before they can be released to the public at all, including tests by third parties and national security experts in Government.”

STATEMENT OF DARIO AMODEI, CHIEF EXECUTIVE OFFICER, ANTHROPIC, SAN FRANCISCO, CALIFORNIA Mr. Amodei. Chairman Blumenthal, Ranking Member Hawley, and ... Second, we recommend a testing and auditing regime for new and more powerful models. Similar to cars or airplanes, AI models of the near future will be powerful machines that possess great utility but can be lethal if designed incorrectly or misused. New AI models should have to pass a rigorous battery of safety tests before they can be released to the public at all, including tests by third parties and national security experts in Government.

His statement on the Paris AI Action Summit (11 February 2025) says “greater focus and urgency is needed” on democracies keeping the lead, the risks of AI and the economic transitions ahead, and points to the Anthropic Economic Index, released that week.

However, greater focus and urgency is needed on several topics given the pace at which the technology is progressing. The need for democracies to keep the lead, the risks of AI, and the economic transitions that are fast approaching—these should all be central features of the next summit. [...] That’s why this week we released the Anthropic Economic Index, which tracks the distribution of economic activities for which people are currently using our AI systems, including whether they augment or automate current human tasks.

The Paris statement also says: “At the next international summit, we should not repeat this missed opportunity.”

At the next international summit, we should not repeat this missed opportunity.

In “Statement from Dario Amodei on the Paris AI Action Summit” (11 February 2025) he wrote: “A “country of geniuses in a datacenter” could represent the largest change to the global labor market in human history. A first step is to monitor and observe the economic impacts of today’s AI systems.”

A “country of geniuses in a datacenter” could represent the largest change to the global labor market in human history. A first step is to monitor and observe the economic impacts of today’s AI systems.

He wrote that Anthropic supported California’s SB 53, which requires the largest AI developers to make their frontier model safety protocols public and exempts companies with annual gross revenue below $500 million.

We therefore supported a carefully designed bill in California where most of America’s leading AI labs are headquartered, including Anthropic. This bill, SB 53, requires the largest AI developers to make their frontier model safety protocols public and is written to exempt any company with an annual gross revenue below $500M—therefore only applying to the very largest AI companies.

He wrote the statement after “a recent uptick in inaccurate claims about Anthropic's policy stances”, among them that Anthropic was interested in harming start-ups and that its models are “uniquely politically biased”.

Despite our track record of communicating frequently and transparently about our positions, there has been a recent uptick in inaccurate claims about Anthropic's policy stances. ... Some have suggested that we are somehow interested in harming the startup ecosystem. ... Some have claimed that Anthropic's models are uniquely politically biased.

In “A statement from Dario Amodei on Anthropic’s commitment to American AI leadership” (21 October 2025) he wrote: “There are products we will not build and risks we will not take, even if they would make money.”

There are products we will not build and risks we will not take, even if they would make money.

In a statement under his name he said two uses had never been in Anthropic’s contracts with the Department of War and should not be added now (mass domestic surveillance; fully autonomous weapons), and that Anthropic “cannot in good conscience accede to their request”.

In Good Company, Norges Bank Investment Management’s podcast, published the episode with him on 26 June 2024 (4,034 seconds, 67 minutes).

In Good Company is hosted by Nicolai Tangen, CEO of Norges Bank Investment Management. New episode out every Wednesday. The podcast In Good Company is part of Norges Bank Investment Management

The episode’s description says In Good Company is hosted by Nicolai Tangen, CEO of Norges Bank Investment Management, and that Amodei “joins Nicolai in Oslo”.

How much bigger and more powerful will the next AI models be? Anthropic's CEO, Dario Amodei, joins Nicolai in Oslo to discuss the latest advancements in AI, the economic impact of this technology, and the importance of responsible scaling. ... In Good Company is hosted by Nicolai Tangen, CEO of Norges Bank Investment Management.

Lex Fridman Podcast #452 with him was published on 11 November 2024 (5 hours 15 minutes).

Economist Enterprise – Events published “Navigating a world in transition: Dario Amodei in conversation with Zanny Minton Beddoes” on 27 January 2025 (45 minutes).

The first Economic Index report, based on millions of anonymised Claude.ai conversations, found AI use leaned towards augmentation (57%) over automation (43%).

Anthropic and Accenture announced a partnership on embedded evaluation, led by Faculty, Accenture’s specialist AI business; each expects to invest at least $1 billion in building capacity in this area over the next five years.

Partnering with Accenture on embedded evaluation. Sep 18, 2026. We're partnering with Accenture on independent evaluation of frontier AI. ... The partnership will be led by Faculty, Accenture’s specialist AI business ... Anthropic and Accenture each expect to invest at least $1 billion in building capacity in this area over the next five years.

Anthropic calls the Accenture partnership non-exclusive, says it will work with other evaluators to be announced, and says it “will fund Accenture's work directly”.

Given the importance and urgency of this work, Anthropic will fund Accenture's work directly. ... Our partnership is non-exclusive; Anthropic will work with other evaluators to be announced in the coming weeks, and Accenture will work with other AI developers in similar capacities.

Anthropic says the Accenture partnership is “an important step toward the commitment, made in our CEO’s essay “We Must Pace the Frontier,” to embed evaluators within Anthropic”.

We're partnering with Accenture on independent evaluation of frontier AI. This is an important step toward the commitment, made in our CEO’s essay “We Must Pace the Frontier,” to embed evaluators within Anthropic.

His site has a page to subscribe to email alerts about new posts.

His site, darioamodei.com, links to his Google Scholar profile, which describes him as CEO and Co-Founder at Anthropic with a verified anthropic.com email; the list it shows includes duplicate and mis-attributed records, such as one entry for “Language models are few-shot learners” carrying the author list of a different paper.

arXiv lists 15 authors, Dario Amodei among them, for the paper behind the Economic Index, submitted 11 February 2025.

Authors:Kunal Handa, Alex Tamkin, Miles McCain, Saffron Huang, Esin Durmus, Sarah Heck, Jared Mueller, Jerry Hong, Stuart Ritchie, Tim Belonax, Kevin K. Troy, Dario Amodei, Jared Kaplan, Jack Clark, Deep Ganguli [...] by Kunal Handa and 14 other authors [...] [Submitted on 11 Feb 2025]

Economic Index paper: the abstract says it analyses over four million Claude.ai conversations.

We leverage a recent privacy-preserving system to analyze over four million this http URL conversations through the lens of tasks and occupations in the U.S. Department of Labor's O*NET Database.

anthropic.com/claude leads to Claude’s product page.

Sam Altman, Dario Amodei, Yoshua Bengio and Clement Delangue briefed the Security Council at the meeting.

“This Council faces an unprecedented threat,” warned Yoshua Bengio, co-Chair of the UN Independent International Scientific Panel on AI ... “We have a choice in front of us,” said Sam Altman, Chief Executive Officer of OpenAI ... added Dario Amodei, Chief Executive Officer of Anthropic ... The Council also heard from Clément Delangue, Chief Executive Officer of Hugging Face

The UN’s summary says Amodei urged Member States to pursue “narrow” agreements that can enjoy broad support, such as a ban on using AI to make biological weapons, and urged efforts to build evaluation and verification systems and to establish common standards for testing AI models.

Offering suggestions for Council action, he urged Member States to pursue “narrow” agreements that can enjoy broad support, such as a ban on using to AI to make biological weapons. Further, he urged efforts to build evaluation and verification systems that keep pace with AI development and to establish common standards for testing AI models for risks involving a loss of control or misuse of the technology.

Donald Trump posted “White House Accord on Super Intelligence - Joint Commitment on Frontier Responsibilities” on Truth Social on 29 September 2026, as a post with two attached images and no text of its own.

"created_at": "2026-09-29T21:24:58.077Z", "content": "<p></p>", two media_attachments of type image. First image heading: White House Accord on Super Intelligence / Joint Commitment on Frontier Responsibilities

The signature page lists Donald J. Trump, President of the United States, and Sundar Pichai, Google; Dario Amodei, Anthropic; Mark Zuckerberg, Meta; Greg Brockman, OpenAI; Elon Musk, XAI; and Jensen Huang, Nvidia.

Donald J. Trump President of the Unites States [...] Sundar Pichai Google [...] Dario Amodei Anthropic [...] Mark Zuckerberg Meta [...] Greg Brockman OpenAI [...] Elon Musk XAI [...] Jensen Huang Nvidia

The text says “we believe each company should implement the following four layers of controls and audits”.

Therefore, in addition to any other precautions, we believe each company should implement the following four layers of controls and audits:

Layer 3 reads “Partner with an independent external auditor or evaluator to carry out independent assessments of whether the controls, monitoring, and detection are operating as intended.”

(3) Partner with an independent external auditor or evaluator to carry out independent assessments of whether the controls, monitoring, and detection are operating as intended.

The text of the commitment states no penalty or enforcement mechanism.

Over time, it may make sense to codify these steps into laws or regulations. Regardless of whether this is required of companies, we believe that implementing these controls and audits is critical to ensuring a safe future for everyone, and each of our companies are committed to doing this.

The DC Circuit decided the case on 25 September 2026, ruling 2-1 against Anthropic; Judge Katsas wrote for the majority, joined by Judge Rao, and Judge Henderson dissented.

Argued May 19, 2026 Decided September 25, 2026 ... Before: HENDERSON, KATSAS, and RAO, Circuit Judges. Opinion for the Court filed by Circuit Judge KATSAS. Dissenting opinion filed by Circuit Judge HENDERSON. ... We reject these challenges. ... Accordingly, we deny the petitions for review.

The court held that the Department had ample support for its conclusion that the continued integration of Claude into its information systems, by the Department or its contractors, presented a statutorily covered national-security risk.

We reject these challenges. The Department had ample support for its conclusion that the continued integration of Claude into the Department's information systems, by the Department or its contractors, presented a statutorily covered national-security risk.

Anthropic calls Claude “our AI assistant trained with Constitutional AI”.

GOV.UK says the AI Safety Summit 2023 would “take place on the 1 and 2 November at Bletchley Park, Buckinghamshire”.

In a statement dated 5 March 2026 he wrote that on 4 March Anthropic received a letter from the Department of War confirming it had been designated a supply chain risk to America’s national security; that “we see no choice but to challenge it in court”; and that for customers it applies only to the use of Claude “as a direct part of contracts with the Department of War”.

The September 2025 report found that 77% of API transcripts showed automation patterns against 12% augmentation, and that on Claude.ai the split was nearly even.

The March 2026 report says automation “decreased sharply” in its first-party API data.

Anthropic publishes the index’s data at anthropic.com/economic-index, last updated 26 June 2026.

Anthropic said on 21 July 2025 that it “intends to sign the European Union's General-Purpose AI Code of Practice”.

Deep reinforcement learning from human preferences: the abstract says the approach solved complex tasks, including Atari games, from human preferences between pairs of trajectory segments, with feedback on less than one percent of the agent’s interactions with its environment.

Anthropic says it is a Delaware Public Benefit Corporation.

Delaware law says a public benefit corporation “shall be managed in a manner that balances the stockholders’ pecuniary interests, the best interests of those materially affected by the corporation’s conduct, and the public benefit or public benefits identified in its certificate of incorporation”.

Annex III lists among high-risk AI systems those “intended to be used to evaluate the creditworthiness of natural persons or establish their credit score, with the exception of AI systems used for the purpose of detecting financial fraud”.

Executive Order 14347 says the Department of Defense may be referred to as the Department of War in official correspondence, public communications and other contexts it lists.

CBS Sunday Morning: “Extended interview: Dario Amodei”, with Jo Ling Kent, published 13 September 2026, 1,427 seconds (24 minutes).

Claude channel: “A conversation with Dario Amodei & Daniela Amodei”, moderated by Ami Vora at the Code with Claude developer conference, published 6 May 2026, 1,990 seconds (33 minutes).

Accenture and Anthropic announced on 9 December 2025 the Accenture Anthropic Business Group, making Anthropic one of Accenture’s select strategic partners “with a dedicated practice built around Claude”.

Photo of Dario Amodei at 10 Downing Street, 24 May 2023, by Simon Walker / No 10 Downing Street, CC BY 2.0 on Commons.

Official White House photograph by Lawrence Jackson, 4 May 2023: Vice President Kamala Harris meets AI company chief executives in the Roosevelt Room; the caption names Sam Altman, Dario Amodei, Satya Nadella and Sundar Pichai as the chief executives present.

Photograph by Simon Walker for No 10 Downing Street, 24 May 2023: Rishi Sunak meets Dario Amodei, Demis Hassabis and Sam Altman. Positions: our identification; the No 10 caption gives none. Flickr shows CC BY-NC-ND 2.0 today; Commons’ review recorded CC BY 2.0 on 12 June 2023.

Prime Minister’s Office of Japan, 29 October 2025: Prime Minister Sanae Takaichi receives a courtesy call from Dario Amodei. Positions: our identification; six people stand in the photo.