This morning I woke up

@kuanning.tseng

8.2K followers · usually 1.0K views a post

Views

10.0K

Likes

560

Comments

SoundTikTok sound · original sound - kuanning.tseng

Why it worked

The video effectively explains a complex geopolitical and technological event (US government action against AI models) in an accessible way, using clear analogies and visual aids. It taps into the current interest and concern surrounding AI development and regulation, making it highly relevant and shareable.

Summary

The video explains that the US government has shut down Anthropic's newest AI models, Fable 5 and Mythos 5, due to an export control directive citing national security. The presenter clarifies that this action, while presented as a security measure, effectively blocks all non-US citizens, including those within the US, and highlights the concentration risk for businesses relying on a single AI model.

Structure

1Introduction to the shutdown of Anthropic's AI models
2Explanation of the government's export control directive
3The impact on US citizens and the broader implications
4The concept of 'jailbreaking' and the government's concern
5The reality of cybersecurity work and Anthropic's stance
6The shift of AI from software to national power and the resulting risks

Product placement

Anthropic's AI models (Fable 5 and Mythos 5) are the subject of the video, with news articles and internal documents about them displayed.

On-screen text

The Washington Post Anthropic shuts down newest AI model after U.S. bans foreign use The company said it faced a sweeping federal export restriction over hacking concerns, posing a national security risk. Today at 12:54 p.m. EDT Make an informed preference on Google AI Announcements Statement on the US government directive to suspend access to Fable 5 and Mythos 5 Jun 12, 2024 WHAT WE'LL DEEP DIVE Three things, in plain English 01 What these two models actually do 02 Why an order aimed at foreign nationals hit everyone 03 Why it matters for you THE WASHINGTON POST THE REALITY CHECK Three days, launch to shutdown JUN 9 JUN 12 - 11:21 AM ET THAT NIGHT Anthropic gives public, most powerful commercial model Fable 5 and Mythos 5 US export control directive issued. Using national security. Both models go dark for all customers. THE WASHINGTON POST Claude Fable 5 and Claude Mythos 5 WHAT HAPPENED What the order actually said An export control directive, citing national security The same legal tool used for weapons and advanced technology. 9-11:21 AM ET the order landed Only US citizens allowed; every non-citizen blocked It reaches foreign nationals even sitting inside the United States, including US staff. So Anthropic shut both models off for all customers Unable to filter for Citizenship at icons, even US citizens not access. All other Anthropic models still work Source: The Washington Post, The Verge THE REALITY CHECK Not the superweapon it sounds like The flaws it found were minor and already known Finding and fixing flaws is daily defender work Sounds harmless, but finding a flaw is the same skill as finding a stay in. Other leading AI models already do the same thing Anthropic is complying, but disagrees: recalling a commercial model over a narrow "jailbreak" would halt new releases. THE REALITY CHECK export control directive apps THE SIGNAL software NATIONAL POWER COMPANY vs COMPANY YOUR STACK 1 MODEL single point of failure CONCENTRATION RISK

Original caption

The US government just made Anthropic switch off its two newest AI models. Overnight. For everyone. The reason is not what the headlines say? 美國政府讓 Anthropic 一夜之間關掉兩個最新的 AI 模型，所有人都不能用。真正的原因是什麼？ #AIeverywhere #AINews #Anthropic #AISecurity #TechTok

More from @kuanning.tseng

More breakout videos from this creator.

Transcript

This morning I woke up and I couldn't believe what just happened because the US government just forced Anthropic to switch off two of its newest AI models, overnight. And the real reason is not what the headlines say. In this video, I'm going to talk about what these two models can do, why an order targeted at foreign nationals hit everyone, ended up shutting them off for everyone, even in America. And why this matters to all of us in the world. So a quick breakdown of how fast this moved. On June 9th, Anthropic launched Fable 5 to the public. Its most powerful commercial model yet, wrapped in heavy safety filters. One tester said it can do two months of engineering work in just a single day. The other model, its twin, Mythos 5, went only to a small group of trusted cyber defenders, never to the public. Anthropic calls it the strongest cybersecurity model in the world. Three days later, on June 12th, the government order hit. By that night, which is last night, both were switched off for everyone. I was actually so sad because in the morning, I was actually pretty excited to wake up and build the MVP, and also check out all of the agents I've built so far. But all of a sudden, it just doesn't work anymore. Here is what happened. The order was an export control directive, citing national security. The rule was really blunt, only US citizens are allowed. Every non-citizen is blocked, even someone sitting inside the United States. So to comply, Anthropic actually shut them down, both of the models, for all customers. So even US citizens right now lose the access. Their other models still work. Here is the reality check. Fable is built to refuse anything that smells like hacking. Ask it to break into any system and it will shut you down. A jailbreak is a way around that block, which is what the US government is worried about. Imagine you're in a club with a bouncer at the front door, but you really want to go in. So you sneak in through the side. So what's the trick here? You don't ask the model to hack. You ask it to read your code and fix some weak spots. But finding the weak spots in the code is the same skill as finding a way in. The model quietly does the work it was built to refuse. Sounds alarming, and here's the catch. Finding and fixing flaws is what cybersecurity teams are actually doing. Anthropic itself is saying that is nothing special. Other leading AI models already do the same thing. Now, the reality check. A government just pulled a commercial AI product offline with an export control. That is a toolkit used for missiles and advanced technology, not for apps. The signal is pretty clear. AI is no longer just software. It's national power. And the race is not just company against company anymore. And nations are starting to wall their AI off from the world. What does that mean for you? If your business runs on a single AI model, one order like this can pull it out from under you overnight. That is the concentration risk. And most teams are not ready for this.