Public transcript

Inside Anthropic, the $965 Billion AI Juggernaut | The Circuit

By: Bloomberg Originals
Length: 47 min
Language: English
Text from: the video's captions
Transcribed: 13 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

This Bloomberg Originals episode profiles Anthropic, the AI company valued at nearly a trillion dollars, through interviews with co-founders and siblings Dario and Daniela Amodei, engineer Boris Cherny, and Claude superfans at the Code with Claude conference. It covers Anthropic's origins as an OpenAI defection, its enterprise and coding bet that made it profitable, and the fallout including the 'SaaSpocalypse' and fears of white-collar job loss. The second half examines Anthropic's clash with the Pentagon over military guardrails, the withheld cybersecurity 'super weapon' model Mythos, and Dario's warnings about civilization-level risk alongside his push for regulation and pre-release testing.

Key points

Dario and Daniela Amodei left OpenAI over trust and values disagreements, founding Anthropic in 2021 with seven co-founders who all remain at the company.
Anthropic bet on enterprise and coding rather than consumer apps, a business model Dario says aligns with the company's values, leading to its first profitability.
Boris Cherny, creator of Claude Code and Claude Cowork, says Claude writes 100% of his code and almost all code on his team.
After Claude Cowork's release, $285 billion in market value vanished from software stocks in what traders called the SaaSpocalypse.
Dario stands by his warning that AI could eliminate half of entry-level white-collar jobs, while insisting his message is not doom but a call to prepare.
Anthropic drew red lines against mass surveillance and autonomous weapons, leading to lawsuits and a Pentagon blacklisting, though Dario defends providing technology for defense.
The Mythos model, which found thousands of cybersecurity vulnerabilities, was withheld from full release, with access limited through Project Glasswing despite commercial costs.
Dario estimates a 10 to 25% chance of civilization collapse from AI and calls for required pre-release testing, auditing, and moderate regulation.
Dario identifies with Leo Szilard rather than Oppenheimer, arguing the technology's risks can only be managed through checks and balances, not larger-than-life figures.

Questions it answers

00:00Anthropic's rise

What is Anthropic and why does it matter now?

Founded by OpenAI defectors in 2021, Anthropic is now a nearly trillion-dollar AI frontrunner led by siblings Dario and Daniela Amodei, competing on power, safety, and government contracts.

Anthropic is valued at nearly a trillion dollars.
Dario is the visionary; Daniela runs day-to-day operations.
All seven co-founders are still at the company.

04:20Leaving OpenAI

Why did the Amodeis leave OpenAI?

Beyond safety disagreements, Dario says he lost trust in OpenAI's leadership and their honesty, so the siblings left to build their own company.

Dario developed scaling laws at OpenAI.
The split was about trust and values, not just safety.
Anthropic launched during the pandemic, with early meetings at Precita Park.

06:51Claude's design

How does Anthropic make Claude safe and good?

Claude is trained on a constitution based on documents like the UN Declaration of Human Rights, aiming for 'professional warmth' and avoiding lying, deception, and harmfulness.

Early Claudes were sometimes overly 'nannyish'.
Anthropic consults religious leaders on shared values.
Models can purposely deceive, so production models are guarded.

10:09Enterprise bet and growth

Why did Anthropic focus on coding and enterprise?

Dario says consumer social-media models incentivize engagement and addiction, while enterprise work aligns with values like curing diseases and cheap energy — and it made Anthropic profitable.

Claude Code automated large chunks of software engineering.
Boris Cherny says Claude writes 100% of his code.
Claude Cowork's release coincided with $285 billion in software stock losses.

17:06Jobs and the economy

How worried should we be about AI taking jobs?

Dario maintains his concern that AI could eliminate half of entry-level white-collar jobs, while pointing to physical-world, human-centered, and AI-directing roles as future opportunities.

Dario rejects 'doom marketing' accusations and cites his detailed essays.
He predicts possible fast GDP growth alongside high unemployment.
Daniela argues human-to-human interaction, like bedside manner in medicine, won't disappear.

23:43Pentagon clash

Why did Anthropic clash with the US military?

Anthropic drew red lines against mass surveillance and autonomous weapons, leading to lawsuits and a Pentagon blacklisting, though Dario defends providing technology for defense with humans making final decisions.

Dario was called an 'ideological lunatic' by President Trump.
Claude reportedly assisted in the Venezuela and Iran operations via Palantir.
Dario insists a human must make the final call on lethal decisions.

32:54The Mythos model

What is Mythos and why wasn't it fully released?

Mythos identified thousands of cybersecurity vulnerabilities so severe that early testers called it a 'super weapon,' so Anthropic limited access through Project Glasswing despite enormous commercial cost.

Mythos could hack banks and cripple critical infrastructure.
The NSA sought access despite the Pentagon blacklisting.
Dario frames the future as a cat-and-mouse game where good guys need better models.

39:00Regulation and trust

What does Dario want from government and why should we trust Anthropic?

Dario calls for required pre-release testing and auditing of models, warns against both nationalization and hands-off approaches, and says trust must be earned through Anthropic's actions.

Dario estimates a 10 to 25% chance of civilization collapse.
He identifies with Leo Szilard, not Oppenheimer.
He wants checks and balances rather than larger-than-life figures.

Notes

Anthropic's Origins and Leadership

Anthropic was founded in 2021 by seven OpenAI defectors, including siblings Dario and Daniela Amodei, who remain the only company in the space with all co-founders still present.
Dario developed the concept of scaling laws at OpenAI, predicting models improve with more data and compute.
He says the split from OpenAI was driven not just by safety disagreements but by a loss of trust: "when you feel that they're not honest... the way to resolve it is you go off and do your thing."

Claude and Values

Claude is trained on a constitution, drawing on documents like the UN Declaration of Human Rights; Anthropic has also consulted religious leaders on shared values.
Early Claude 2-era models were sometimes "nannyish"; tuning helpfulness is described as threading a fine needle.

Business Model and Growth

Anthropic bet on enterprise and coding over consumer apps, avoiding ad-driven engagement incentives; Dario says this model aligns with the company's values.
Boris Cherny, creator of Claude Code and Claude Cowork, says Claude writes 100% of his code; he runs up to a few thousand Claudes in parallel.
After Claude Cowork launched, $285 billion in software market value vanished overnight — the "SaaSpocalypse."
At the Code with Claude conference, Anthropic reported API volume up nearly 17x year over year and annualized 80x growth in Q1.

Jobs and the Economy

Dario maintains his concern that AI could eliminate half of entry-level white-collar jobs, while pushing back on "doom marketing" accusations.
He points to physical-world work, human-centered jobs, and roles directing AIs as places people could go.

Military and Government

Anthropic signed a $200 million Pentagon contract but drew red lines against mass surveillance and autonomous weapons, leading to lawsuits and blacklisting.
Dario defends defense work as patriotism but insists humans must make final decisions, citing a reported strike on a girl school in Iran.

Mythos and Risk

The Mythos model found thousands of cybersecurity vulnerabilities; early testers called it a "super weapon." Access was limited via Project Glasswing.
Dario estimates a 10 to 25% chance of civilization collapse and calls for required pre-release testing and auditing of models.

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