The AI industry moves so fast that if you blink, you miss a major shift. This past week was one of the most chaotic we’ve seen in a while—from leaked internal models to massive stock market crashes caused by a single algorithm.
Here is the breakdown of the 5 biggest AI updates you need to know about.
1. Anthropic Accidentally Leaks “Claude Mythos” Anthropic made a massive, embarrassing mistake. They accidentally left over 3,000 internal documents wide open on the internet. Inside, researchers found traces of a secret, unreleased model called Claude Mythos (internally called Capybara).
Why does this matter? Anthropic claims it’s a massive step up in coding and cybersecurity. However, the hilarious irony? Anthropic stated the model is so good at finding security holes that they are terrified of their own creation.
2. Google Drops “TurboQuant” — And Chip Stocks Crashed For years, the assumption has been that AI needs more hardware. Well, Google Research just dropped an algorithm called TurboQuant. It compresses AI memory by 6 times and speeds up processing by 8 times—without losing accuracy.
The market realized: If software can make AI this efficient, do we actually need to spend billions building new chip factories? The news hit Wall Street, and chip stocks like Micron, Samsung, and SK Hynix dropped 5-6% almost instantly.
3. OpenAI Officially Kills “Sora” OpenAI’s AI video generator, Sora, is officially dead. It burned hundreds of millions in GPU costs but only made $2 million total. Why did it fail? Because according to a leaked memo from Sam Altman, Sora was blocking the path to their real goal: AGI deployment. The Sora team has been quietly moved to work on AI for robots instead.
4. Claude Can Now Use Your Computer Remotely (Claude Dispatch) This is the most practical update for power users. Anthropic released Claude Dispatch. You can now send a text message from your phone while you’re at dinner: “Export my pitch deck from my laptop to a PDF.” Claude will spin up your computer, open the file, and email it to your client. You don’t have to touch your keyboard.
5. Luma AI Drops “Uni-1” & MiroFish Simulates 1000 Opinions In the image generation space, Luma released Uni-1, a model that thinks and creates at the same time. Instead of generating a generic image, it figures out the actual physical structure and measurements of the object first.
Meanwhile, a 20-year-old built MiroFish, an open-source tool that simulates thousands of different perspectives on a decision to help you see where people agree or disagree.
Conclusion From leaked models to market-crashing algorithms, the infrastructure of AI is shifting under our feet. If you want to keep up with what actually matters—and ignore the hype—this is your weekly digest.
Which of these 5 updates surprised you the most? Let me know in the comments!