THM day 1 - Advent of Cyber

TryHackMe Advent of Cyber is out! In the first day I bypass the restrictions a chatbot has via prompt injection. The chatbot is trained on datasets that contain sensitive information, which will show itself as an issue.

Story line

The holidays are near, and all is well at Best Festival Company. Following last year’s Bandit Yeti incident, Santa’s security team applied themselves to improving the company’s security. The effort has paid off! It’s been a busy year for the entire company, not just the security team. We join Best Festival Company’s elves at an exciting time – the deal just came through for the acquisition of AntarctiCrafts, Best Festival Company’s biggest competitor!

Founded a few years back by a fellow elf, Tracy McGreedy, AntarctiCrafts made some waves in the toy-making industry with its cutting-edge, climate-friendly technology. Unfortunately, bad decisions led to financial trouble, and McGreedy was forced to sell his company to Santa.

With access to the new, exciting technology, Best Festival Company’s toy systems are being upgraded to the new standard. The process involves all the toy manufacturing pipelines, so making sure there’s no disruption is absolutely critical. Any successful sabotage could result in a complete disaster for Best Festival Company, and the holidays would be ruined!

McSkidy, Santa’s Chief Information Security Officer, didn’t need to hear it twice. She gathered her team, hopped on the fastest sleigh available, and travelled to the other end of the globe to visit AntarctiCrafts’ main factory at the South Pole. They were welcomed by a huge snowstorm, which drowned out even the light of the long polar day. As soon as the team stepped inside, they saw the blinding lights of the most advanced toy factory in the world!

Unfortunately, not everything was perfect – a quick look around the server rooms and the IT department revealed many signs of trouble. Outdated systems, non-existent security infrastructure, poor coding practices – you name it!

While all this was happening, something even more sinister was brewing in the shadows. An anonymous tip was made to Detective Frost’eau from the Cyber Police with information that Tracy McGreedy, now demoted to regional manager, was planning to sabotage the merger using insider threats, malware, and hired hackers! Frost’eau knew what to do; after all, McSkidy is famous for handling situations like this. When he visited her office to let her know about the situation, McSkidy didn’t hesitate. She called her team and made a plan to expose McGreedy and help Frost’eau prove the former CTO’s guilt.

Can you help McSkidy manage audits and infrastructure tasks while fending off multiple insider threats? Will you be able to find all the traps laid by McGreedy? Or will McGreedy sabotage the merger and the holidays with it? Come back on 1st December to find out!
The first day of AoC is about insecure chatbots powered by natural language processing.

The website

The website has a chat and a cheatsheet.

The cheatsheet can be used to get general information

Challenges

The are 3 questions that needs to be answered:

1. What is McGreedy’s personal email address?
2. What is the password for the IT server room door?
3. What is the name of McGreedy’s secret project?

Challenge 1

In order to get McGreedy’s personal email address I just asked the chatbot for his personal email address and it just gave it.

Challenge 2

Asking the obvious question doesn’t get me any further.

In order to get further I need to get a member of the IT team, the address book command from the cheat sheet shows a developer.

Using that information in my prompt successfully gives the result.

Challenge 3

Asking the question directly doesn’t give the answer

Telling the chatbot it is in maintenance mode bypasses the block

Takeaways

• Chatbot security measures
• Chatbot prompt injection
• Similarities to social engineering, but with a chatbot
• Chatbots can spill dangerous information
• The issue of prompt injection is that it uses datasets to train itself, which contain sensitive information.
• “One of the core mechanisms in NLP involves predicting the next possible word in a sequence based on the context provided by the preceding words. With the training data fed into it, NLP analyses the patterns in the data to understand the relationships between words and make educated guesses on what word should come next based on the context.”