🧠 Anthropic claims DeepSeek stealing Claude's Capabilities

Good Morning! Anthropic just accused Chinese AI labs of misusing Claude to rapidly train their own models, raising fresh concerns around AI security and model distillation. Plus, you will learn how to reverse-engineer and generate viral short-form videos using AI

Plus, in today’s AI newsletter:

• Anthropic Just Caught DeepSeek Stealing Claude's Capabilities

• Single-Model AI Chips Go Ultra-Fast

• Citrini Research Warns Scariest Economic Crisis

• How to recreate viral YouTube shorts without filming

• 4 new AI tools worth trying

AI RACE

🧠 Anthropic Just Caught DeepSeek Stealing Claude's Capabilities

Anthropic claims DeepSeek, MiniMax, and Moonshot ran large-scale operations to extract Claude’s capabilities, allegedly bypassing safeguards and accelerating their model development.

• Around 24,000 fake accounts and 16M+ interactions were used to access Claude at scale

• DeepSeek allegedly targeted Claude’s reasoning skills and generated “censorship-safe” political answers

• Anthropic warns distilled models may lack safety guardrails and be used for surveillance or cyber ops

• OpenAI previously made similar accusations, calling it “free-riding” on U.S. AI labs

This highlights how frontier AI models can be copied faster and cheaper than building from scratch. As AI becomes geopolitically sensitive, model misuse and distillation could become the next major battleground.

AI HARDWARE

⚡ Single-Model AI Chips Go Ultra-Fast

Taalas unveiled HC1, a custom chip designed to run only one AI model, ditching general-purpose hardware entirely to maximize speed and efficiency.

• The chip permanently embeds Llama 3.1 8B into hardware, delivering responses in under 100 ms

• Taalas claims ~100× faster output than standard hardware, with far lower power and cost

• While the model is older and non-frontier, new chips for stronger models can be built in months

• The startup raised $169M this round ($200M+ total), with a mid-size reasoning model chip coming soon

If this “model-baked-into-silicon” approach scales to frontier systems, it could unlock real-time AI for robotics, physical AI, and agentic workflows where every millisecond counts.

AI RESEARCH

🧭 Citrini Research Warns Scariest Economic Crisis

Citrini Research released a speculative memo set in 2028 that explores how runaway AI success could trigger a systemic economic crisis driven by white-collar displacement and collapsing demand.

• Mass adoption of agentic AI boosts productivity, cuts white-collar jobs, and initially lifts corporate margins and stocks

• Consumer spending falls as displaced workers stop spending, creating “Ghost GDP”, output that doesn’t circulate in the real economy

• SaaS margins compress fast, not because AI replaces software, but because buyers believe it can and negotiate harder

• Biggest risk zones: private credit tied to SaaS growth and the $13T mortgage market built on stable white-collar incomes

The memo highlights a dangerous feedback loop where AI success erodes the demand side of the economy. For builders and investors, the real question is no longer if AI works, but what breaks when human intelligence is no longer the scarce input.

HOW TO AI

🗂️ How to recreate viral YouTube shorts without filming

In this tutorial, you will learn how to reverse-engineer and generate viral short-form videos using AI tools like NotebookLM, OpenArt, and ElevenLabs, allowing you to produce high-quality, consistent content without ever picking up a camera.

🧰 Who is This For

• Content creators looking to scale their channel without hiring actors

• Researchers and marketers analyzing and replicating proven viral video formulas

• Faceless channel owners who need consistent, cinematic AI video generation

• Anyone who wants AI to handle scripting, storyboarding, and voiceovers

STEP 1: Analyze & Script

Find a YouTube channel that is crushing it with a specific style, and copy the links to 3 to 6 of their best-performing Shorts.

Head to Google NotebookLM, create a new notebook, and paste those links as your sources. In the chat box, give it a detailed prompt describing what is happening: “This is a handheld vlog style POV where a wife records her husband's daily interactions with a wolf. Create a similar 30-second short script with 10 scenes.” 

Hit Enter, and NotebookLM will analyze the pacing, tone, and structure to build your script.

STEP 2: Generate Consistent Visuals with OpenArt

Now we need visuals. The biggest mistake beginners make is using "text-to-video," which often results in random, inconsistent clips. We will use "image-to-video" for maximum control.

Ask NotebookLM to generate detailed image prompts for every scene in your new script. Then go to OpenArt.ai, select "Images," and choose a realistic model (like Nano Banana Pro). Set your aspect ratio to vertical (9x6 or 9x16) for Shorts.

Generate your first image for scene one.

Once you have an image you like, use OpenArt's "Reference Image" feature and upload that first picture. Keep generating the rest of your scenes using this reference to ensure your characters, lighting, and environments look perfectly consistent.

STEP 3: Animate the Images into Video

Go back to NotebookLM and ask it to generate "video animation prompts" (like camera movements or subtle motions) for all your images.

In OpenArt, switch to the "Image to Video" section and select a top-tier video model. Upload your generated images one by one, paste the corresponding motion prompt, and set the duration to about 4 seconds.

Use the first-frame and last-frame control features to ensure smooth, natural transitions between your clips, avoiding the stiff, robotic look common in AI videos.

STEP 4: The Final Polish (Voice & Edit)

You have silent video clips. Now we need to add the emotion and assemble the final product.

Take your finished script and paste it into ElevenLabs to generate a realistic, natural-sounding voiceover (keep the speed at 1x to maintain authenticity). Finally, open a free editor like CapCut.

Import your AI video clips and your audio track, arranging everything so the narration syncs perfectly with the visuals. Add a few subtle sound effects, like footsteps or wind, to build immersion, then click export. Your video will be ready to post.

OpenAI launches Frontier Alliances, entering into multiyear deals with Accenture, BCG, Capgemini, and McKinsey to help deploy its enterprise platform Frontier.

A DOD official says xAI has agreed to let the military use Grok in classified systems and agreed to the “all lawful use” standard, which Anthropic has refused.

Anthropic has lined up $5B to $6B for an employee share sale at a valuation of about $350B; outside investors will be buying up the insider shares.

Tencent closed its TiMi Montréal studio after nearly five years, without ever releasing a game, as Chinese giants scale back funding for Western game studios.

💻 Claude in PPT: Anthropic’s AI that helps you build PowerPoint slides

📸 Pomelli: Turns product photos into ready-to-use marketing content

🎥 Replit Animation: Turn text prompts into animated videos

⚙️ Rork Max: AI tool to build native iOS apps

THAT’S IT FOR TODAY

Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!

See you tomorrow :)

- Dr. Alvaro Cintas