RabbitOS Intern: Weaponizing AI for Twitter OSINT

I like to use technology in unconventional ways because the hacker in me still enjoys pushing boundaries until something breaks. Always on the hunt for the best affordable OSINT tools, I recently stumbled upon an unsuspecting superweapon: the Rabbit R1.

I’ve covered the Rabbit R1 previously, but not like this.

Despite the negative feedback from people who don’t see the value in this handy little orange PTT device, half the size of an iPhone, it has been instrumental in giving me incredible leverage when analyzing large datasets from scrapes related to threat intelligence.

This all stemmed from my research and experiments using ChatGPT for intelligence analysis. The results were mind-blowing, but it may surprise you that I found the rabbit’s customized analysis abilities, powered by rabbitOS Intern, more autonomous.

In contrast, I had to hold ChatGPT’s hand throughout the entire process, and my master OSINT query list didn’t flow as autonomously as I hoped. This means I had to run every prompt in my OSINT framework manually, which was time-consuming, even though the results were amazing.

Compared to OpenAI’s leading global artificial intelligence (AI) model, if you’ve already built your own OSINT framework, such as a list of prompts you want AI to use when analyzing data and convert into a report, then hands down, RabbitOS’s Intern has demonstrated clear superiority.

Who knew?

Let me show you how it’s done.

Twitter analysis on the go

There are a variety of ways I gather data, especially since I’m often AFK (Away From Keyboard).

You might ask, “What about your phone?” No, I hate typing on it, and I don’t want to see another notification or get distracted by anything.

If you own an R1 device, you gain access to a generous range of free, useful services, many of which traditional subscription-based platforms would charge you for. As for me, I can perform a variety of OSINT tasks directly from it, anywhere, anytime.

For example, you can analyze X (formerly Twitter) posts, aside from simply posting to your own timeline, which feels more like a novelty if you're only using it for trivial tasks.

Since it doesn’t utilize X’s public API, the R1 uses what’s called the ‘Cookie Jar’ from the rabbithole to capture and store session cookies. These cookies are temporarily kept in an encrypted vault within a disposable virtual environment in the cloud, without storing your username or password.

For R1 users:

1. Log in to the rabbithole
2. Click on ‘Cookie Jar’ in the upper right corner
3. Click ‘Save new cookies’
4. Enter the URL of the site you want to save cookies for

A virtualized browser will appear, showing the site you entered. This is where you enter your credentials, thus creating the session cookie. Then hit the orange button at the bottom right corner, ‘I logged in.’’

Doing this will open a new range of options and voice prompts from the R1.

Yes, you can now post tweets directly on your timeline from the R1, just by saying something like, “Please tweet, ‘How is everyone doing today?’” But don’t get sidetracked!

Next, I asked the R1 to summarize my own X posts, because I wanted to see how useful it would be when launched in the wild as an OSINT tool. Its response was helpful, but a little too generalized for a person who’s used to processing oceans of information.

Results are not only shown on the R1, but are saved and timestamped in the rabbithole, under ‘Journal’, which also provides a download feature so you can save your results as a text document.

Its response was factual, but lacked contextual insight, which is where ChatGPT excels. However, I see this improving in the future, since Rabbit Inc. is constantly evolving the capabilities of its AI model.

In an effort to expand my R1’s capabilities, I used teach mode, which is designed to let users train their R1 in order to automate tasks. I taught it some basic OSINT queries, but since teach mode is still in beta, the results were inconsistent.

I can hear some snide chuckles in the background somewhere. I’m not done yet.

While I’m on an OSINT warpath that will surely break my R1 in the process, the results I can produce using the rabbitOS Intern are completely superior to the ChatGPT 4o AI model.

ZenScraper and rabbitOS power combo

Let me introduce you to ZenScraper, a powerful and versatile Twitter scraper that doesn’t rely on public API keys.

Instead, it uses an authentication session cookie to access X content for scraping. Elegantly written by 0Day3xpl0it, the tool was inspired by a desire to provide a “practical solution to aid investigators in collecting crucial evidence” for online child safety investigators.

0Day3xpl0it told Cybernews, “ZenScraper was created to better understand X's API and GraphQL systems, moving away from slower scraping methods like HTML, CSS, and DOM.”

He explained that, given his own previous experience working with OpChildSafety, the inspiration to build an efficient, free OSINT tool for civilian-led investigators was important to him.

“It's recommended to use a backup account that's actively used to build some credibility on X and prevent your main account from getting flagged due to excessive scraping,” he added.

I decided to scrape my last 100 tweets, although the tool is brilliant at also scraping retweets, the bio, or scraping it all in one go. I saved the output to export1.txt.

How to Install:

git clone https://github.com/0Day3xpl0it/zenscraper.git

cd zenscraper

chmod +x *.py

pip install -r requirements.txt

playwright install

Next, generate an authenticated session cookie:

python3 grab_x_cookies.py

A browser will spawn, prompting you to log in to your X account. The script creates an x_cookies.json file, which is mandatory for authenticated scraping.

Using rabbitOS Intern for AI analysis

Before I begin, let me mention that R1 users are afforded nine free daily credits to use rabbitOS Intern, while non-R1 users receive three, as part of its current free trial phase. It will eventually become a paid subscription service available to everyone.

In late April, I was contacted by Matthew Domko, Head of Security at Rabbit Inc., who increased my daily RabbitOS Intern credits to 30 to support my ongoing research with Intern.

Now it was time to analyze export1.txt with the rabbitOS Intern using my full OSINT framework for performing a full intelligence analysis of the tweets. For every dataset I have uploaded to Intern since discovering its capabilities, it has proven its ability to autonomously perform complicated tasks. Better yet, because I own an R1, this function is free to use.

You see, rabbitOS Intern incorporates both contextual awareness and pattern recognition analysis as vital components of its AI-driven functions. The difference between Intern and ChatGPT is that ChatGPT was primarily designed for conversational interactions and provides responses based on user input without task automation.

However, Intern goes beyond mere conversation by integrating autonomous task execution and multi-agent coordination. This means it can perform complex sequences of actions without direct input from the user. While I use both platforms in the field when I’m conducting OSINT investigations, I now use them for different purposes.

Because I gave Intern 24 points to research, it took about 45 minutes for it to complete its analysis, but the wait is worth it. It produced a whopping 15-page analysis report with an insane level of detail, carrying out the objective in ways that exceeded my expectations.

Not only did it analyze my tweets, but it also extended its investigation beyond a simple scrape, into the wider web, gathering public information about me from various news sources and constructing a terrifyingly accurate portrait of its understanding of the subject in question.

This included a heatmap derived from the timestamps of my posts, mentions of accounts I interact with, and, most unsettlingly, an embarrassing amount of information about my past criminal hacking charge that occurred in my mid-20s, which sent me to prison for 11 years, even generating images to illustrate aspects of the crime.

Furthermore, Intern created a diagram illustrating a potential attack surface related to the crime, accurately capturing the factual narrative: that I once worked as a security guard and acted as an insider threat at a healthcare facility, using the elevator to access various floors containing computer systems.

I should have unplugged my computer from the wall, because the wealth of information was pretty personal, but now I had a dialogue box open with Intern, which meant we could discuss the findings further.

It was waiting for my response. Here, we had an in-depth discussion about its findings, which were far from the generalized responses I experienced earlier using the cookie jar and teach mode.

I will probably delete the report.

Not because it wasn’t useful. It was a work of art. After all, same as if you hired a qualified intern to conduct a 24-point OSINT analysis investigation well — that’s why it’s called Intern.

Now you see why I prefer rabbitOS Intern over ChatGPT as an apex OSINT tool.

Always remember to independently verify information when using AI models for research.