Public transcript

Watch: Anthropic CEO Dario Amodei From World Economic Forum | WSJ

By: The Wall Street Journal
Length: 32 min
Language: English
Text from: the video's captions
Transcribed: 12 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

At a WSJ Journal House event in Davos, Anthropic CEO Dario Amodei argues that businesses and governments are not prepared for AI's economic impact, which he says could bring high GDP growth alongside high unemployment and inequality. He defends Anthropic's enterprise-focused strategy and safety record, discusses competition with Google and OpenAI, and fields reader questions on interpretability, education, and AI's effect on developing countries. He predicts that concerns about distributing AI's gains will become bipartisan and universal within a couple of years.

Key points

Amodei says AI may bring 5-10% GDP growth together with 10% unemployment and high inequality, a combination societies have never faced.
He argues the Anthropic Economic Index is a necessary first step because policy will be blind without measuring the economic transition.
Anthropic's enterprise-focused business model lets it avoid consumer-AI incentives around engagement, ads and slop.
Revenue grew from roughly $100 million in 2023 to roughly $1 billion in 2024 to roughly $10 billion in 2025.
Claude Co-work, built in about a week and a half almost entirely with Claude Opus, saw metrics about four times higher than any previous release.
He sees mechanistic interpretability as the key missing breakthrough for making AI reliably safe and controllable.
He warns AI is uniquely suited to strengthening autocracies through surveillance, propaganda and drone armies.
He predicts concerns about distributing AI's gains will become bipartisan and universal within a year or two.

Questions it answers

00:00Opening and AI's impact debate

Are businesses and governments doing enough to prepare for AI's impact?

No, says Amodei. He argues the technology's capabilities have advanced on a constant exponential curve while public perception has swung wildly, and he sees no awareness of the magnitude of what is coming, especially economically.

Amodei says public opinion oscillates while the technology follows a smooth exponential 'Moore's law for intelligence'.
He believes AI could bring both high GDP growth and high unemployment, a combination never seen before.

06:14Adapting to economic disruption

How could society organize itself to adapt to high GDP growth with high unemployment?

Amodei outlines three steps: measure the transition via the Anthropic Economic Index, help people adapt (likely toward physical-world jobs), and accept a government role in distributing the gains. He says the priority should shift from worrying about disincentivizing growth to ensuring everyone shares in it.

The Anthropic Economic Index tracks what Claude is used for, automation vs augmentation, and diffusion across regions.
Amodei expects more jobs in the physical world and fewer in knowledge work.
He says government will need a role in distributing the larger economic pie.

09:39Engaging with the administration

Is Amodei talking to the US administration about these concerns?

Yes, he has said these things to them himself and will talk to officials at Davos. He notes Anthropic agreed with most of the administration's AI action plan but wants a public debate rather than to control policy.

Anthropic agreed with the vast majority of the administration's AI action plan.
He says Anthropic's role is to describe what it sees and provide data, not drive policy.

11:00Safety, competition and China

Have competitive pressures compromised Anthropic's safety principles?

Amodei says Anthropic's enterprise focus reduces the need to fight its own business incentives, and it has made sacrifices like aggressive safety testing and pioneering mechanistic interpretability. On China, he frames the concern as about autocracy, not competition, arguing AI could enable surveillance, individualized propaganda and drone armies.

The enterprise business model avoids consumer-AI incentives around engagement and ads.
Anthropic tests for deception, blackmail and sycophancy and pioneered mechanistic interpretability.
He argues targeted policies like not selling chips matter more than fighting adversaries.

15:12Claude's breakout moment and business growth

How is Anthropic's business doing and what about an IPO?

Amodei describes revenue growing on a smooth exponential curve, from roughly $100 million in 2023 to roughly $10 billion in 2025, with breakout moments around Claude Code and Claude Co-work. On an IPO, he says they don't know for sure, but notes the industry's high capital demands exceed what private markets can provide indefinitely.

Revenue roughly went from zero to $100M in 2023, to $1B in 2024, to $10B in 2025.
Claude Co-work was built in about a week and a half almost entirely with Claude Opus.
He says the industry has high capital demands that private markets can only supply for so long.

20:00Competing with Gemini and media generation

Does Amodei worry about competing with Google's Gemini?

He says being different helps: Google and OpenAI are fighting over consumer, which is existential to both, while Anthropic focuses on enterprise. On lacking video and photo generation, he says enterprise demand doesn't require it and Anthropic could contract a model from someone else if needed.

Google and OpenAI are fighting over consumer, which is existential to both.
He sees little enterprise demand for generated photos or video.

22:09Scientists vs entrepreneurs, geopolitics

How do scientist-led AI companies differ from entrepreneur-led ones?

Amodei says scientists have a long tradition of taking responsibility for the effects of their technology, whereas the social-media entrepreneur generation was shaped by different selection effects and consumer manipulation. On US-EU tensions, he says Anthropic is independent, focused on AI, and has seen no reluctance to work with it.

Scientists have a tradition of taking responsibility for the technology they build.
He says Anthropic keeps its focus on AI and speaks for itself on policy.

25:33Reader questions

What did Amodei say about interpretability, education, and countries being left behind?

He calls mechanistic interpretability the key missing breakthrough for safe AI. On education, he says cheating is a short-term issue and advocates returning to education that builds character rather than purely economic skills. On inequality, he warns of a decoupled 'zeroth world' and points to Anthropic's work with Rwanda's Ministry of Education and the Gates Foundation, predicting these concerns will become bipartisan.

Mechanistic interpretability is 'the only ground truth we have' for model safety.
He favors education aimed at character and enrichment rather than purely economic outcomes.
He warns of a decoupled 'zeroth world' and says government involvement will become bipartisan.

Notes

AI's economic impact

Amodei says AI could produce a combination never seen before: 5-10% GDP growth alongside 10% unemployment and high inequality.
He compares AI progress to a "Moore's law for intelligence" — a smooth exponential curve — while public opinion oscillates between hype and dismissal.
AI coding is a microcosm: some Anthropic engineering leads no longer write code, just edit Claude's output. Software may become cheap or essentially free, but whole careers built for decades may disappear.
He sees no awareness of "the magnitude of what is coming."

Preparing society

Step one: the Anthropic Economic Index, updated four or five times over about a year, tracks what Claude is used for, automation vs augmentation, and diffusion across states and countries. "Until we can measure the shape of this economic transition, any policy is going to be blind."
Step two: help people adapt — likely more jobs in the physical world, fewer in knowledge work; robotics is on a slower trajectory.
Step three: government will need a role in distributing the larger pie; he says it's time to worry less about disincentivizing growth.

Anthropic's strategy and safety

The enterprise focus avoids consumer-AI incentives (engagement maximization, slop, ads).
Anthropic tests models for deception, blackmail and sycophancy, and pioneered mechanistic interpretability — "the only ground truth we have" for safety.
On China, the concern is autocracy, not competition: AI could enable surveillance, individualized propaganda, breaking into computer systems, and drone armies. Targeted policies like not selling chips matter most.

Business

Revenue: roughly zero to $100M (2023), $100M to $1B (2024), $1B to $10B (2025).
Breakout moments: Claude Code among developers with Opus 4.5, and Claude Co-work, built in about a week and a half almost entirely with Claude Opus for non-coders; its metrics were about four times anything previously released.
On an IPO: undecided, but private markets can only supply so much capital.

Education and inequality

He advocates education that builds character and enriches rather than a purely economic, "mercenary" notion.
Nightmare scenario: a decoupled "zeroth world" of ~10 million people centered on Silicon Valley capturing most growth.
Anthropic works with Rwanda's Ministry of Education and the Gates Foundation; he predicts these distribution concerns will become bipartisan within a year or two.

Text from the video's own captions. Summary and notes written by AI from the transcript, so check anything important against the recording.