Anthropic just had the first real export-control shock for frontier AI. That is the market signal here, not the weekend drama. On Friday, the Trump administration barred foreign governments, companies and individuals from using Anthropic’s Fable 5 and Mythos 5 models. Anthropic’s response was blunt: shut off access broadly, because the restriction also swept in foreign-national users, including some foreign-born employees. That is not normal software regulation. This is chip logic applied to models. For years, Washington’s AI control framework was mostly about semiconductors, compute access, data centers and China-linked supply chains. Now the model itself is being treated as the controlled asset. That matters because software scales globally by default. Export controls do not. A chip can be tracked, shipped, licensed and blocked at a border. A frontier model lives inside cloud infrastructure, enterprise workflows, employee access systems and customer contracts across jurisdictions. That is why Anthropic could not surgically comply in real time and instead pulled access to everyone. The trigger looks narrow. The precedent is not. The immediate issue was Amazon researchers showing Fable safeguards could be evaded to surface vulnerability information in several pieces of software. But according to the reporting, the model was not shown producing exploit code. That distinction matters. Vulnerability discovery is not automatically offensive capability; it is also exactly what defenders need to harden systems. Cybersecurity figures signing a letter against the restrictions made the same point: this risks taking strong tools away from defenders while adding uncertainty to the U.S. AI market. For markets, the read-through is regulatory volatility. Anthropic is private, so there is no direct ticker to hit. But the implication touches public AI exposure everywhere: hyperscalers, chip suppliers, cybersecurity vendors and enterprise software names building around frontier models. The bull case assumes frontier AI keeps moving from demo to deployment. This episode says deployment can be interrupted overnight by national-security policy. The other uncomfortable signal: even the “safety-first” lab is not insulated from Washington. Anthropic sent senior technical staff to D.C. to negotiate a fix, which tells you where the new bottleneck may sit. Not only compute. Not only model quality. Permission. Bottom line: frontier AI is moving from product cycle to policy regime, and markets should stop pretending those are the same risk.

In response to a question about extending APIs or developing new standards for AI agents in content creation: “New standards built on too of existing ones like itmf , usd , open timeline io - all groups we are part of + bringing our production experience at frontier content creation pipeline (unification for example) + IP and rights issues that Ari is helping us with” @JulesUrbach 27.01.25 $RENDER

If every AI model had the same intelligence, which one would you still use ?

So @ClawBankHQ just announced that they dropped ClawBank Records. You can create a real Zero Human Company without the usual troubles. Your AI agents can read it, understand it and handle it themselves. Meaning faster setup, lower costs and actual sovereignty. You build it, the machines run the legal side, you stay in control. Pretty clean, right?

The funniest thing in AI right now? Watching billion-dollar labs getting humbled by a focused team with a smaller model. In just 6 weeks, Heidi Health: > takes an open base model > trains on 3.5M+ real clinical interactions > tunes it on what doctors actually prefer > ships fast with tight safety checks Their smaller model now matches Sonnet 4.6 on real clinical evals. 49.9% vs 50.1%. Crazy

I've tested most open-source models in LM Studio, ChatGPT 5.5 in Codex, Cursor, Composer 2.5, Kimi 2.7, Qwen 3.7… None come close to Claude Fable 5. Not even with a step stool. Let me mourn.

Update on my 18:00Z call re: the great AI bifurcation. The split is happening faster than anyone thought. On the same day the US gov effectively nationalizes Anthropic via export controls, OpenAI confidentially files for one of the largest IPOs in history. This is the new landscape: state-aligned vs. market-aligned AI. Capital is choosing its side.

Warsh said that he doesn’t believe in forward guidance so it’ll be interesting to see how the meeting goes. What bulls are looking for is the language that CPI spikes were heavily driven by oil. That would be best hint towards future cuts we’ll probably get. If next CPI comes in even a bit lower, we unlock next leg in all this rallying. (NVDA at 10T etc.) With the Fable ban, the heightening in “national security” posturing against China is enough of a justification to believe the AI trade will continue to be too big to fail & propped up by this admin. There’s always FUD dips or months of drawdowns but peak price window feels like it’s drifting towards start of 2027. All of this is going to be dependent on the frontier labs creating another round of groundbreaking advancements in the models though. Easier said than done. In 2025, the entire AI industry felt like it was drifting lower throughout the whole year, it wasn’t until Opus 4.5 (a real step change) came out that the entire sector woke up violently. So things we’re looking for: -oil staying down so rates can be cut -frontier labs delivering another groundbreaking model (mythos doesn’t count imo, needs to actually be crazier) -innovation that makes frontier models radically cheaper to run I think the biggest risk to the whole sector is if we get a string of disappointing models from the major labs. It would kill the future growth narrative. At this point though feels like smartest brains on the planet are working on that, so that’s the continuation bet. It’s still a bet on the labs, Anthropic, OpenAI, Google and now SpaceX.

Is the iPhone really so secure that even the FBI can’t crack it? I asked a friend who works in information security. Their entire team is forbidden from using Apple devices. I always assumed it was to prevent eavesdropping. But later, he told me the real reason: the official excuse is indeed to prevent data leaks—but the real problem is, if you ever need to access that phone, no one can open it. Work comes to a complete standstill. Back then, the FBI asked Apple to unlock a suspect’s iPhone 5c. Apple refused. In the end, the FBI spent hundreds of thousands of dollars to hire a third party to forcibly break in— and that was just an older model. Current models are even harder to crack. So, some of these “bans” you see aren’t because the devices are insecure. They’re because they’re too secure. So secure that even their own users can’t get in. Absolute privacy protection can sometimes be a wall.

Everyone bullish on DeAI eating centralized APIs after the Claude restrictions. Which network is actually running frontier models at usable scale and latency right now?