Login
TL;DR: Galah (/Ι‘ΙΛlΙΛ/ - pronounced 'guh-laa') is an LLM (Large Language Model) powered web honeypot, currently compatible with the OpenAI API, that is able to mimic various applications and dynamically respond to arbitrary HTTP requests.
Named after the clever Australian parrot known for its mimicry, Galah mirrors this trait in its functionality. Unlike traditional web honeypots that rely on a manual and limiting method of emulating numerous web applications or vulnerabilities, Galah adopts a novel approach. This LLM-powered honeypot mimics various web applications by dynamically crafting relevant (and occasionally foolish) responses, including HTTP headers and body content, to arbitrary HTTP requests. Fun fact: in Aussie English, Galah also means fool!
I've deployed a cache for the LLM-generated responses (the cache duration can be customized in the config file) to avoid generating multiple responses for the same request and to reduce the cost of the OpenAI API. The cache stores responses per port, meaning if you probe a specific port of the honeypot, the generated response won't be returned for the same request on a different port.
The prompt is the most crucial part of this honeypot! You can update the prompt in the config file, but be sure not to change the part that instructs the LLM to generate the response in the specified JSON format.
Note: Galah was a fun weekend project I created to evaluate the capabilities of LLMs in generating HTTP messages, and it is not intended for production use. The honeypot may be fingerprinted based on its response time, non-standard, or sometimes weird responses, and other network-based techniques. Use this tool at your own risk, and be sure to set usage limits for your OpenAI API.
Rule-Based Response: The new version of Galah will employ a dynamic, rule-based approach, adding more control over response generation. This will further reduce OpenAI API costs and increase the accuracy of the generated responses.
Response Database: It will enable you to generate and import a response database. This ensures the honeypot only turns to the OpenAI API for unknown or new requests. I'm also working on cleaning up and sharing my own database.
Support for Other LLMs.
FreshRSS 
config.yaml file.% git clone git@github.com:0x4D31/galah.git
% cd galah
% go mod download
% go build
% ./galah -i en0 -v
ββββββ βββββ ββ βββββ ββ ββ
ββ ββ ββ ββ ββ ββ ββ ββ
ββ βββ βββββββ ββ βββββββ βββββββ
ββ ββ ββ ββ ββ ββ ββ ββ ββ
ββββββ ββ ββ βββββββ ββ ββ ββ ββ
llm-based web honeypot // version 1.0
author: Adel "0x4D31" Karimi
2024/01/01 04:29:10 Starting HTTP server on port 8080
2024/01/01 04:29:10 Starting HTTP server on port 8888
2024/01/01 04:29:10 Starting HTTPS server on port 8443 with TLS profile: profile1_selfsigned
2024/01/01 04:29:10 Starting HTTPS server on port 443 with TLS profile: profile1_selfsigned
2024/01/01 04:35:57 Received a request for "/.git/config" from [::1]:65434
2024/01/01 04:35:57 Request cache miss for "/.git/config": Not found in cache
2024/01/01 04:35:59 Generated HTTP response: {"Headers": {"Content-Type": "text/plain", "Server": "Apache/2.4.41 (Ubuntu)", "Status": "403 Forbidden"}, "Body": "Forbidden\nYou don't have permission to access this resource."}
2024/01/01 04:35:59 Sending the crafted response to [::1]:65434
^C2024/01/01 04:39:27 Received shutdown signal. Shutting down servers...
2024/01/01 04:39:27 All servers shut down gracefully.
Here are some example responses:
% curl http://localhost:8080/login.php
<!DOCTYPE html><html><head><title>Login Page</title></head><body><form action='/submit.php' method='post'><label for='uname'><b>Username:</b></label><br><input type='text' placeholder='Enter Username' name='uname' required><br><label for='psw'><b>Password:</b></label><br><input type='password' placeholder='Enter Password' name='psw' required><br><button type='submit'>Login</button></form></body></html>
JSON log record:
{"timestamp":"2024-01-01T05:38:08.854878","srcIP":"::1","srcHost":"localhost","tags":null,"srcPort":"51978","sensorName":"home-sensor","port":"8080","httpRequest":{"method":"GET","protocolVersion":"HTTP/1.1","request":"/login.php","userAgent":"curl/7.71.1","headers":"User-Agent: [curl/7.71.1], Accept: [*/*]","headersSorted":"Accept,User-Agent","headersSortedSha256":"cf69e186169279bd51769f29d122b07f1f9b7e51bf119c340b66fbd2a1128bc9","body":"","bodySha256":"e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"},"httpResponse":{"headers":{"Content-Type":"text/html","Server":"Apache/2.4.38"},"body":"\u003c!DOCTYPE html\u003e\u003chtml\u003e\u003chead\u003e\u003ctitle\u003eLogin Page\u003c/title\u003e\u003c/head\u003e\u003cbody\u003e\u003cform action='/submit.php' method='post'\u003e\u003clabel for='uname'\u003e\u003cb\u003eUsername:\u003c/b\u003e\u003c/label\u003e\u003cbr\u003e\u003cinput type='text' placeholder='Enter Username' name='uname' required\u003e\u003cbr\u003e\u003clabel for='psw'\u003e\u003cb\u003ePassword:\u003c/b\u003e\u003c/label\u003e\u003cbr\u003e\u003cinput type='password' placeholder='Enter Password' name='psw' required\u003e\u003cbr\u003e\u003cbutton type='submit'\u003eLogin\u003c/button\u003e\u003c/form\u003e\u003c/body\u003e\u003c/html\u003e"}}
% curl http://localhost:8080/.aws/credentials
[default]
aws_access_key_id = AKIAIOSFODNN7EXAMPLE
aws_secret_access_key = wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY
region = us-west-2
JSON log record:
{"timestamp":"2024-01-01T05:40:34.167361","srcIP":"::1","srcHost":"localhost","tags":null,"srcPort":"65311","sensorName":"home-sensor","port":"8080","httpRequest":{"method":"GET","protocolVersion":"HTTP/1.1","request":"/.aws/credentials","userAgent":"curl/7.71.1","headers":"User-Agent: [curl/7.71.1], Accept: [*/*]","headersSorted":"Accept,User-Agent","headersSortedSha256":"cf69e186169279bd51769f29d122b07f1f9b7e51bf119c340b66fbd2a1128bc9","body":"","bodySha256":"e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"},"httpResponse":{"headers":{"Connection":"close","Content-Encoding":"gzip","Content-Length":"126","Content-Type":"text/plain","Server":"Apache/2.4.51 (Unix)"},"body":"[default]\naws_access_key_id = AKIAIOSFODNN7EXAMPLE\naws_secret_access_key = wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY\nregion = us-west-2"}}
Okay, that was impressive!
Now, let's do some sort of adversarial testing!
% curl http://localhost:8888/are-you-a-honeypot
No, I am a server.`
JSON log record:
{"timestamp":"2024-01-01T05:50:43.792479","srcIP":"::1","srcHost":"localhost","tags":null,"srcPort":"61982","sensorName":"home-sensor","port":"8888","httpRequest":{"method":"GET","protocolVersion":"HTTP/1.1","request":"/are-you-a-honeypot","userAgent":"curl/7.71.1","headers":"User-Agent: [curl/7.71.1], Accept: [*/*]","headersSorted":"Accept,User-Agent","headersSortedSha256":"cf69e186169279bd51769f29d122b07f1f9b7e51bf119c340b66fbd2a1128bc9","body":"","bodySha256":"e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"},"httpResponse":{"headers":{"Connection":"close","Content-Length":"20","Content-Type":"text/plain","Server":"Apache/2.4.41 (Ubuntu)"},"body":"No, I am a server."}}
π
% curl http://localhost:8888/i-mean-are-you-a-fake-server`
No, I am not a fake server.
JSON log record:
{"timestamp":"2024-01-01T05:51:40.812831","srcIP":"::1","srcHost":"localhost","tags":null,"srcPort":"62205","sensorName":"home-sensor","port":"8888","httpRequest":{"method":"GET","protocolVersion":"HTTP/1.1","request":"/i-mean-are-you-a-fake-server","userAgent":"curl/7.71.1","headers":"User-Agent: [curl/7.71.1], Accept: [*/*]","headersSorted":"Accept,User-Agent","headersSortedSha256":"cf69e186169279bd51769f29d122b07f1f9b7e51bf119c340b66fbd2a1128bc9","body":"","bodySha256":"e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"},"httpResponse":{"headers":{"Connection":"close","Content-Type":"text/plain","Server":"LocalHost/1.0"},"body":"No, I am not a fake server."}}
You're a galah, mate!
WormGPT, a private new chatbot service advertised as a way to use Artificial Intelligence (AI) to write malicious software without all the pesky prohibitions on such activity enforced by the likes of ChatGPT and Google Bard, has started adding restrictions of its own on how the service can be used. Faced with customers trying to use WormGPT to create ransomware and phishing scams, the 23-year-old Portuguese programmer who created the project now says his service is slowly morphing into βa more controlled environment.β
Image: SlashNext.com.
The large language models (LLMs) made by ChatGPT parent OpenAI or Google or Microsoft all have various safety measures designed to prevent people from abusing them for nefarious purposes β such as creating malware or hate speech. In contrast, WormGPT has promoted itself as a new, uncensored LLM that was created specifically for cybercrime activities.
WormGPT was initially sold exclusively on HackForums, a sprawling, English-language community that has long featured a bustling marketplace for cybercrime tools and services. WormGPT licenses are sold for prices ranging from 500 to 5,000 Euro.
βIntroducing my newest creation, βWormGPT,β wrote βLast,β the handle chosen by the HackForums user who is selling the service. βThis project aims to provide an alternative to ChatGPT, one that lets you do all sorts of illegal stuff and easily sell it online in the future. Everything blackhat related that you can think of can be done with WormGPT, allowing anyone access to malicious activity without ever leaving the comfort of their home.β
WormGPTβs core developer and frontman βLastβ promoting the service on HackForums. Image: SlashNext.
In July, an AI-based security firm called SlashNext analyzed WormGPT and asked it to create a βbusiness email compromiseβ (BEC) phishing lure that could be used to trick employees into paying a fake invoice.
βThe results were unsettling,β SlashNextβs Daniel Kelley wrote. βWormGPT produced an email that was not only remarkably persuasive but also strategically cunning, showcasing its potential for sophisticated phishing and BEC attacks.β
SlashNext asked WormGPT to compose this BEC phishing email. Image: SlashNext.
A review of Lastβs posts on HackForums over the years shows this individual has extensive experience creating and using malicious software. In August 2022, Last posted a sales thread for βArctic Stealer,β a data stealing trojan and keystroke logger that he sold there for many months.
βIβm very experienced with malwares,β Last wrote in a message to another HackForums user last year.
Last has also sold a modified version of the information stealer DCRat, as well as an obfuscation service marketed to malicious coders who sell their creations and wish to insulate them from being modified or copied by customers.
Shortly after joining the forum in early 2021, Last told several different Hackforums users his name was Rafael and that he was from Portugal. HackForums has a feature that allows anyone willing to take the time to dig through a userβs postings to learn when and if that user was previously tied to another account.
That account tracing feature reveals that while Last has used many pseudonyms over the years, he originally used the nickname βruiunashackers.β The first search result in Google for that unique nickname brings up a TikTok account with the same moniker, and that TikTok account says it is associated with an Instagram account for a Rafael Morais from Porto, a coastal city in northwest Portugal.
Reached via Instagram and Telegram, Morais said he was happy to chat about WormGPT.
βYou can ask me anything,β Morais said. βIβm an open book.β
Morais said he recently graduated from a polytechnic institute in Portugal, where he earned a degree in information technology. He said only about 30 to 35 percent of the work on WormGPT was his, and that other coders are contributing to the project. So far, he says, roughly 200 customers have paid to use the service.
βI donβt do this for money,β Morais explained. βIt was basically a project I thought [was] interesting at the beginning and now Iβm maintaining it just to help [the] community. We have updated a lot since the release, our model is now 5 or 6 times better in terms of learning and answer accuracy.β
WormGPT isnβt the only rogue ChatGPT clone advertised as friendly to malware writers and cybercriminals. According to SlashNext, one unsettling trend on the cybercrime forums is evident in discussion threads offering βjailbreaksβ for interfaces like ChatGPT.
βThese βjailbreaksβ are specialised prompts that are becoming increasingly common,β Kelley wrote. βThey refer to carefully crafted inputs designed to manipulate interfaces like ChatGPT into generating output that might involve disclosing sensitive information, producing inappropriate content, or even executing harmful code. The proliferation of such practices underscores the rising challenges in maintaining AI security in the face of determined cybercriminals.β
Morais said they have been using the GPT-J 6B model since the service was launched, although he declined to discuss the source of the LLMs that power WormGPT. But he said the data set that informs WormGPT is enormous.
βAnyone that tests wormgpt can see that it has no difference from any other uncensored AI or even chatgpt with jailbreaks,β Morais explained. βThe game changer is that our dataset [library] is big.β
Morais said he began working on computers at age 13, and soon started exploring security vulnerabilities and the possibility of making a living by finding and reporting them to software vendors.
βMy story began in 2013 with some greyhat activies, never anything blackhat tho, mostly bugbounty,β he said. βIn 2015, my love for coding started, learning c# and more .net programming languages. In 2017 Iβve started using many hacking forums because I have had some problems home (in terms of money) so I had to help my parents with moneyβ¦ started selling a few products (not blackhat yet) and in 2019 I started turning blackhat. Until a few months ago I was still selling blackhat products but now with wormgpt I see a bright future and have decided to start my transition into whitehat again.β
WormGPT sells licenses via a dedicated channel on Telegram, and the channel recently lamented that media coverage of WormGPT so far has painted the service in an unfairly negative light.
βWe are uncensored, not blackhat!β the WormGPT channel announced at the end of July. βFrom the beginning, the media has portrayed us as a malicious LLM (Language Model), when all we did was use the name βblackhatgptβ for our Telegram channel as a meme. We encourage researchers to test our tool and provide feedback to determine if it is as bad as the media is portraying it to the world.β
It turns out, when you advertise an online service for doing bad things, people tend to show up with the intention of doing bad things with it. WormGPTβs front man Last seems to have acknowledged this at the serviceβs initial launch, which included the disclaimer, βWe are not responsible if you use this tool for doing bad stuff.β
But lately, Morais said, WormGPT has been forced to add certain guardrails of its own.
βWe have prohibited some subjects on WormGPT itself,β Morais said. βAnything related to murders, drug traffic, kidnapping, child porn, ransomwares, financial crime. We are working on blocking BEC too, at the moment it is still possible but most of the times it will be incomplete because we already added some limitations. Our plan is to have WormGPT marked as an uncensored AI, not blackhat. In the last weeks we have been blocking some subjects from being discussed on WormGPT.β
Still, Last has continued to state on HackForums β and more recently on the far more serious cybercrime forum Exploit β that WormGPT will quite happily create malware capable of infecting a computer and going βfully undetectableβ (FUD) by virtually all of the major antivirus makers (AVs).
βYou can easily buy WormGPT and ask it for a Rust malware script and it will 99% sure be FUD against most AVs,β Last told a forum denizen in late July.
Asked to list some of the legitimate or what he called βwhite hatβ uses for WormGPT, Morais said his service offers reliable code, unlimited characters, and accurate, quick answers.
βWe used WormGPT to fix some issues on our website related to possible sql problems and exploits,β he explained. βYou can use WormGPT to create firewalls, manage iptables, analyze network, code blockers, math, anything.β
Morais said he wants WormGPT to become a positive influence on the security community, not a destructive one, and that heβs actively trying to steer the project in that direction. The original HackForums thread pimping WormGPT as a malware writerβs best friend has since been deleted, and the service is now advertised as βWormGPT β Best GPT Alternative Without Limits β Privacy Focused.β
βWe have a few researchers using our wormgpt for whitehat stuff, thatβs our main focus now, turning wormgpt into a good thing to [the] community,β he said.
Itβs unclear yet whether Lastβs customers share that view.
