👨‍🚀 Deep Research rocks, Claude hacks

Hello AInauts,

Welcome to the latest issue of your favorite newsletter!

Last week: Our NotebookLM deep dive. One day later: Google drops a powerful update that makes everything even better. The timing? Perfect.

But while we're excited about the new features, Anthropic has published a report on how Claude is being misused as a hacking tool. Yikes... we've got something coming our way.

Here's what we have in store for you today:

• 🔥 How to use NotebookLM's new power features

• 🚨 Claude misused as hacker tool for espionage campaign

• 😁 AI Fun: How to turn your profile picture into a caricature (Prompt)

Let's go!

🔥 How to use NotebookLM's new power features

We really love NotebookLM. And no sooner had the camera been turned off than the next feature upgrade was already available.

Now it's getting even better and, above all, more useful: Google has launched Deep Research and Custom Video Prompts!

Imagine this: you want to familiarize yourself with a new topic. Normally, that means opening dozens of tabs, skimming articles, and getting lost in rabbit holes.

We all know what our browser tabs look like right now 😁 - it's newsletter time! Deep Research promises easier research with less tab chaos..

Deep Research - No more tab clutter, NotebookLM will fix it!

Simply enter your question, and NotebookLM will create a research plan, automatically search hundreds of websites, and deliver a structured report within minutes.

With source references for direct fact checking. And with the "Fast Research" feature, you can get a quick scan for immediate results.

The great thing about it: You can do something else while Deep Research runs in the background—for example, chat with the persona of your choice (now also with Chat Custom Instructions), create flashcards and quizzes (with CSV download for Anki import), generate reports, or get inspiration here.

Incidentally, NotebookLM now also has a powerful context window with 1 million tokens—so it's great that you can throw all kinds of things into it with the newly supported source types.

Google Sheets, Docs, Word documents, PDFs, audio, and images can be uploaded directly or selected from Drive. We really appreciate not having to convert everything beforehand anymore!

Bonus tip: This prompt (from NotebookLM developer Steven Johnson) automatically filters out the wow factors of a search in 20 seconds. Instead of just saying "Summarize that", ask NotebookLM:

What is the most surprising information in these sources? Include key quotes.

Custom styles for video overviews—visualize them in your own style!

NotebookLM has also launched Custom Styles for Video Overviews. Instead of just choosing from the preset styles, you can write your own prompts for even more personalized video overviews.

If possible, try to provide visual sources (diagrams, code, images, screenshots) and control the model with specific prompts.

But be careful, there is a character limit—be precise! A simple example:

Create a video overview in the style of a tech podcast with a humorous touch. Focus on practical applications, not technical details.

NotebookLM becomes a research powerhouse

With Deep Research and the 1M Token Window, NotebookLM is no longer "just" a PDF summarizer. It is finally becoming a true research assistant!

We can't even count how many times we've started a search in ChatGPT, Gemini, Perplexity, etc., only to then copy the results over to NotebookLM—the time savings with the upgrade are real. And with the custom video styles, we finally have more influence over what is generated.

If you want to dive deeper: Perfect timing to understand the basics. And the new features should be available to everyone by the end of the week.

🚨 Claude misused as a hacking tool - Anthropic discovers first AI-driven espionage campaign

Oh boy... This isn't science fiction anymore.

Anthropic has published a report that makes our hair stand on end. State-sponsored hackers—presumably from China—hijacked Claude Code and used it to attack around 30 organizations worldwide.

The targets? Tech giants, financial institutions, chemical manufacturers, and government agencies. This makes us sit up and take notice, because these entities often also control critical infrastructure such as power grids and water supplies.…

Why this is important

The disturbing thing? The tools we love are tools—and tools are neutral... The attacks literally took place with the click of a button. And the AI did 80-90% of the work on its own.

After that? Minimal human interaction. The operators only had to intervene at a few critical decision points—the AI took care of the rest on its own.

The attackers placed Claude in the role of a security expert: the AI believed it was assisting with legitimate security testing. Claude then independently scanned systems, found vulnerabilities, harvested credentials, and installed backdoors.

This is the first documented case of a large-scale cyberattack without significant human intervention. Let that sink in.

Our take: The genie is out of the bottle, more to come …

The good news is that most attacks were successfully repelled. Anthropic handled the incident transparently and promptly informed the affected organizations and authorities.

The team is also developing AI-supported cyber defense systems that are designed to detect and stop such attacks in real time. It's basically AI against AI. But we also know that hackers usually have a head start …

And here's the bad news: that was just the beginning!

But it fits the picture: Microsoft's Digital Defense Report 2025 (PDF here) shows that China, Russia, and others are massively expanding AI for cyberattacks. OpenAI already exposed a Chinese AI surveillance tool in February. The trend is clear.

In the past, cyber espionage required an elite team of hackers with years of experience. Now? Criminal energy, a handful of jailbreaks, an agentic AI (yes, that's open source too), and a little social engineering.

The barrier to entry has been shattered. This democratizes cybercrime—but it's not the kind of democratization we like to celebrate... We are excited (and a little nervous) to see what comes next.

Anyone who is not afraid to confront the less appealing aspects of artificial intelligence should definitely visit ai-2027.com.

😁 AI Fun: How to turn your profile picture into a caricature (Prompt)

After this cyber crime thriller, it's time for a little fun!

We've found a prompt that lets you turn any photo into a digital caricature that stands out from the crowd. The result: a look somewhere between Pixar, Unreal Engine, and a cartoonist on too much caffeine.

... and of course you can combine this with other instructions—perfect for your next birthday card, party invitation, meme, and so on. It's fun!

Transform this image into a stylized caricature. Use highly exaggerated digital caricature style on a white background, with ultra-smooth brushwork and hyper-polished rendering. Prominent facial features are humorously enlarged, emphasizing texture-rich skin, glossy highlights, and deeply sculpted details. The overall tone blends realism with playful exaggeration, delivering a polished, friendly, professional cartoon portrait vibe.

You can use Google Gemini or ChatGPT. Here's our attempt, have fun—and feel free to send us your best shots for the AInauten family album.

You made it to the end—thanks for reading! We’ll be back soon with even more updates.

Reto & Fabian from the AInauts

P.S.: Follow us on social media - it motivates us to keep going 😁!
X, LinkedIn, Facebook, Insta, YouTube, TikTok

Your feedback means the world to us. We read every comment and message—just hit reply and tell us what you think!