🔴 Anthropic CEO’s urgent AI warning

Good Morning! Anthropic CEO Dario Amodei just published a stark warns that humanity may be closer to real AI danger than we thin, and not ready to handle what’s coming. Plus, you’ll learn how to pull real-time data from any website (even without an API).

In today’s AI newsletter:

• How to Pull Real-Time Data From Any Website (Even Without an API)

• Anthropic CEO Warns We’re Not Ready for Powerful AI

• Qwen3-Max-Thinking Overtakes GPT-5.2 & Gemini

• Anthropic rolls out a new extension to MCP

• 4 new AI tools worth trying

AI IMPACT

⚠️ Anthropic CEO Warns We’re Not Ready for Powerful AI

In a 38-page essay, Amodei outlines the risks of “powerful AI”, systems smarter than the best humans, capable of autonomous action, scientific discovery, and large-scale disruption, arguing society is approaching a critical test moment.

• Claims we’re closer to serious AI risk in 2026 than in 2023, despite the shift to AI optimism

• Warns advanced AI could enable bioweapons, autonomous weapons, mass surveillance, and propaganda

• Predicts up to 50% of entry-level white-collar jobs could disappear within 1–5 years

• Calls for AI “constitutions,” transparency laws, continuous monitoring, and international regulation

As companies race for trillion-dollar upside, Amodei argues that without guardrails, governance, and global coordination, AI’s power could outpace society’s ability to control it.

AI MODELS

🧪 Qwen3-Max-Thinking Overtakes GPT-5.2 & Gemini

Alibaba Cloud’s Qwen team unveiled a proprietary “thinking” model built for agentic workflows, combining iterative reasoning with autonomous tool use, and the benchmarks are loud.

• Scores 49.8 on Humanity’s Last Exam (with search), beating GPT-5.2-Thinking and Gemini 3 Pro

• Uses test-time scaling to iteratively reflect, avoid dead ends, and focus compute where it matters

• Natively integrates search, memory, and code execution inside its thinking loop

• Aggressively priced at $1.20/M input + $6/M output tokens, undercutting most frontier models

Qwen is proving that efficient reasoning + tool autonomy can outperform bigger Western models, while staying cheap enough for real enterprise deployment.

AI NEWS

🧩 Anthropic rolls out a new extension to MCP

Anthropic expanded its open Model Context Protocol (MCP) with MCP Apps, enabling fully interactive apps to run inside Claude instead of returning plain text responses.

• Use Slack, Figma, Canva, Asana, monday.com, and more directly inside Claude

• Interact visually: preview Slack messages, customize Canva decks, explore charts

• Built on MCP, an open standard now backed by OpenAI, Google, Microsoft, and others

• Designed to work beyond Claude, unlocking app-in-chat experiences across AI tools

Healthcare runs on trust, regulation, and accountability. While AI can improve access and understanding, today’s consumer chatbots operate in a gray zone, where your most sensitive data is protected largely by promises, not enforceable safeguards.

HOW TO AI

🧠 How to Pull Real-Time Data From Any Website (Even Without an API)

In this tutorial, you’ll learn how to use Mino, a new automation tool that gives you real-time access to data from websites that don’t offer APIs. Instead of relying on developer endpoints, Mino runs browser-based agents that visit websites like a human would, extract the information you care about, and return it in clean, structured JSON.

🧰 Who is This For

• Founders and operators who want live data from competitor websites

• Developers building automations without relying on APIs

• Analysts doing pricing, product, or market research

• Anyone who wants structured data from “API-less” websites

STEP 1: Create a New Project and Define the Task

Head over to Mino and creating a new project from the dashboard.

To create a new project, locate the “Create Project“ button on top left corner and click it. You’ll immediately notice that everything revolves around a single prompt. In the prompt field, clearly describe what you want the agent to do on a website.

For example, you might write something like:
“Go to the input URL. Find all t-shirts currently in stock. Extract the product name and price. Return the results as structured JSON.”

This prompt is the instruction set your agent will follow every time it runs. You don’t need to think about HTML, selectors, or scraping logic. Just describe the task in plain language.

STEP 2:  Add Website URLs to Run the Agent On

Once the prompt is ready, you pair it with websites. You can either manually enter URLs or upload a list, depending on your use case. Each URL will be treated as a separate job using the same instructions.

For example, you can add Buck Mason, Todd Snyder, and an Italian luxury brand site all at once. The same agent logic runs across all of them, even though their layouts and designs are completely different.

This is where Mino shines. The agent adapts to each site in real time using browser automation instead of rigid rules.

STEP 3: Run the Agent and Watch It Browse the Web

After setting up the URLs, run the job. You’ll see a live preview of the agent interacting with the website like a real user. It scrolls, clicks, searches, and navigates pages to find the information you asked for.

Behind the scenes, the agent is using a real browser session, which is why it works even on sites with no APIs, dynamic content, or complex layouts. The run may take a minute or two depending on the site, but no manual intervention is needed.

STEP 4: Get Structured JSON Results for Automation

When the run finishes, Mino returns clean, structured JSON with the extracted data. You’ll see fields like product name and price neatly organized, ready to be consumed by other tools or workflows.

At this point, you can automate anything. Track prices across retailers, build recommendation systems, monitor inventory changes, or feed the data into downstream automations. You can reuse the same prompt for new URLs, schedule runs, or refine the instructions as your use case evolves.

Once set up, it’s essentially a reusable data agent for the parts of the web that were previously inaccessible.

Microsoft unveils the Maia 200, its 2nd-generation AI accelerator built on TSMC's 3nm process, deploying today in its Azure US Central data center region.

Meta says it will test new paid subscriptions that give people access to exclusive features on its apps and plans to scale Manus as part of those offerings.

A test of ChatGPT Health and Claude for Healthcare with data from Apple Health finds the chatbots provided questionable and inconsistent responses.

📱Clawdbot: Your own 24/7 AI employee

⚙️ Claude Code: Anthropic’s AI coding tool that understands deep context

📊 Claude in Excel: Use Claude to create and edit spreadsheets

📚 Gemini: Personalized SAT prep features powered by AI

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