TubeReads

The US Government Gave Anthropic 90 Minutes to Shut Down Its AI

In a move that shocked Silicon Valley, the US Commerce Department gave Anthropic — a company preparing for a near-trillion-dollar IPO — just 90 minutes on a Friday evening to shut down its flagship AI models. The reason? A supposed security threat that turned out to be the model doing exactly what it was designed to do: read code and fix bugs. But here's where it gets interesting: the whistleblower wasn't a government researcher or independent security expert. It was Amazon CEO Andy Jassy, whose company had invested $13 billion in Anthropic, committed to spending $100 billion on their cloud services, and just happens to be building competing AI models.

Duración del vídeo: 30:09·Publicado 20 jun 2026·Idioma del vídeo: English
6–7 min de lectura·4,446 palabras habladas → resumido a 1,383 palabras (3x)·

1 —

Puntos clave

1

Anthropic received 90 minutes notice to comply with export restrictions that forced them to shut down their flagship models globally because they couldn't verify users' citizenship status in real time.

2

The «jailbreak» that triggered the government intervention was simply asking the AI to read code and fix bugs — the exact functionality the product was designed to provide.

3

Amazon CEO Andy Jassy reported Anthropic to the White House despite Amazon being their largest investor and customer, while simultaneously developing competing AI models.

4

Chinese open-source AI models now dominate usage metrics, processing twice as many tokens as US competitors while costing 60 times less and being immune to US government kill switches.

5

If AI becomes a competitive market rather than a winner-takes-all monopoly, the productivity gains will flow to businesses using the models instead of investors paying near-trillion-dollar valuations.

En resumen

The US government's abrupt shutdown of Anthropic's AI models — triggered by a cloud competitor and based on functionality the product was designed to perform — may have inadvertently accelerated the shift toward cheaper Chinese open-source alternatives, undermining the very monopoly dynamics that justify near-trillion-dollar AI valuations.


2 —

The 90-Minute Shutdown

Commerce Department forced Anthropic to disable flagship AI models with minimal notice.

On a Friday evening, Anthropic received a letter from Commerce Secretary Howard Lutnick giving them 90 minutes to suspend all access to their Fable 5 and Mythos 5 models for any foreign national anywhere in the world. The timing was particularly awkward: Anthropic had recently filed confidential paperwork with the SEC for an IPO disclosing a $47 billion revenue run rate and an expected valuation of $965 billion.

The letter was an «is-informed» notice used in export control cases, directing the company to obtain individual licenses before allowing any foreign person to access the models. It threatened criminal and civil penalties for non-compliance. Because there was no way to instantly verify the citizenship and immigration status of every user, Anthropic had no choice but to disable the models globally for everyone.

The directive applied not just to users in other countries but to any foreign national inside the United States on visas, using a mechanism called «deemed export.» This meant Anthropic was forced to lock its own non-US citizen engineers out of the systems they had just spent the previous year building.


3 —

The «Threat» That Wasn't

The dangerous jailbreak was simply asking the AI to read code.

💡

The «Threat» That Wasn't

According to Anthropic's statement, the technique that caused government panic «essentially consists of asking the model to read a specific code base and fix any software flaws.» In other words, asking a coding AI to read computer code and fix bugs — not a complex cyber weapon exploit, but the product working exactly as advertised.


4 —

Real-World Impact on Cybersecurity

Security professionals lost their best defensive tool overnight.

More than 100 top cybersecurity executives from companies like Google, Zoom, and Sophos signed an open letter to the administration pointing out that the government had just confiscated their best defensive tool. The experts noted that Anthropic had already built significant protections into Fable 5 to prevent offensive hacking use.

Those safety measures were so aggressive that on launch day they became a source of humor in the cybersecurity community. Security researcher Valentina Palmyotti reported the model would refuse to complete innocuous tasks like reading an ordinary IT blog post just in case it contained words vaguely related to computer networks. Cybersecurity veteran Matt Suchi told TechCrunch that if you asked Fable 5 to write secure code, it assumed you were doing dangerous work and immediately downgraded itself to a less advanced model.


5 —

The Amazon Connection

Anthropic's largest investor reported them to the White House.

FINANCIAL RELATIONSHIP
Amazon's $13 Billion Investment
Amazon is one of Anthropic's largest financial backers, having invested $13 billion with commitments to put in up to $25 billion more. Anthropic is also a massive Amazon customer, having committed more than $100 billion to Amazon Web Services to train and run its AI models. This creates an extraordinarily complex partnership where your investor is also your landlord.
COMPETITIVE DYNAMICS
Building Rival Products
Amazon develops its own competing AI models, meaning Andy Jassy's decision to report Anthropic to the White House simultaneously protected the civilized world from automated software debugging while kneecapping his primary corporate rival. When asked for comment, Amazon released a statement saying they don't share details of discussions with governments about potential security risks.

6 —

Dario Amodei's Ironic Timing

CEO advocated for government AI controls one day before shutdown.

“The problem Amodei wrote was that Treebeard operates at a very slow speed compared to hobbits.”

— Dario Amodei


7 —

The G7 Fallout

European leaders panicked about American control over critical AI infrastructure.

A few days after the shutdown, the leaders of the G7 nations gathered in France for a summit that included a special lunch with the «AI3» — Dario Amodei of Anthropic, Sam Altman of OpenAI, and Demis Hassabis of Google DeepMind. The purpose was to discuss AI opportunities for the financial sector, but by the time the soup arrived, JP Morgan had already cut off Anthropic's access for its Hong Kong staff, following a similar move by Goldman Sachs.

French President Emmanuel Macron observed that the Anthropic dispute had clarified the stakes, warning that if the United States can turn off the switch from one day to the next, it would damage the multi-trillion dollar companies leading the AI arms race. He criticized America for interfering too much in business — a remarkable reversal. In response, European politicians proposed the European Technological Sovereignty Package, directing approximately $490 billion toward building their own data centers and AI models.


8 —

Key Financial Metrics

Anthropic's valuation and the competitive landscape in numbers.

Expected IPO Valuation
$965 billion
Up from $380 billion in February, making it one of the largest IPOs in corporate history
Revenue Run Rate
$47 billion
Disclosed in confidential SEC filing according to Financial Times
DeepSeek Cost Advantage
60x cheaper
DeepSeek costs $0.87 per million output tokens versus Anthropic's Fable 5
Chinese Model Market Share
2x tokens processed
Chinese models in the global top 20 process twice as many tokens as US competitors
European Sovereignty Package
€422 billion
Approximately $490 billion directed toward building EU data centers and AI models

9 —

The Chinese Alternative

🎯
Perfectly Timed Launch
Beijing-based AI lab Zhipu announced a new open-source model at exactly 5:21 p.m., a direct reference to the precise minute Anthropic received its shutdown order — corporate pettiness at an international scale.
🔓
Immune to Kill Switches
Because Chinese models like Zhipu and DeepSeek are open source and self-hostable, companies can run them on their own servers, making them entirely immune to being randomly switched off by US authorities.
💰
Radical Cost Advantage
DeepSeek's flagship model costs 87 cents per million output tokens, roughly 60 times cheaper than Anthropic's Fable 5, while still being «good enough» for most business applications.
📊
Market Dominance
In the first two weeks of June, four of the five most popular models on OpenRouter were Chinese, with Chinese models processing twice as many tokens as US competitors in the global top 20.

10 —

The Valuation Problem

Near-trillion-dollar valuations require global monopolies, not competitive markets.

⚠️

The Valuation Problem

To justify a $965 billion valuation, Anthropic must convince investors its total addressable market is essentially everyone on Earth with an internet connection. But if the flagship product is classified as a national security threat requiring background checks, or if open-source alternatives are good enough at 1/60th the cost, that monopoly assumption collapses. The productivity gains everyone was promised will still happen — they'll just flow to businesses using the models rather than investors who paid near-trillion-dollar valuations.


11 —

Valores mencionados

AMZNAmazon

12 —

Personas

Dario Amodei
CEO of Anthropic
mentioned
Donald Trump
President of the United States
mentioned
Howard Lutnick
Commerce Secretary
mentioned
David Sacks
White House AI Adviser
mentioned
Andy Jassy
CEO of Amazon
mentioned
Emmanuel Macron
President of France
mentioned
Sam Altman
CEO of OpenAI
mentioned
Demis Hassabis
CEO of Google DeepMind
mentioned

Glosario
Is-informed letterA private notice the US government uses in export control cases to tell a company it now needs a license for things it didn't need a license for yesterday.
Deemed exportA legal theory where granting a foreign citizen access to controlled technology inside the US is treated the same as shipping it to their home country.
Constitutional AIAnthropic's safety framework designed to build AI systems that politely refuse potentially harmful requests to prevent misuse.
Total addressable market (TAM)The total revenue opportunity available if a product achieves 100% market share in its category.
Open-source modelAn AI model whose code is publicly available and can be run on a company's own servers rather than accessed through a third-party service.

Aviso legal: Este es un resumen generado por IA de un vídeo de YouTube con fines educativos y de referencia. No constituye asesoramiento de inversión, financiero o legal. Verifique siempre la información con las fuentes originales antes de tomar decisiones. TubeReads no está afiliado con el creador de contenido.