Skip to main content

CND News and Blog

Identifying Plagiarism Using AI Bots Like ChatGPT

Featured

For several months we have been seeing a huge interest in the capabilities of ChatGPT and with a high school teacher in the family, we have looked a little deeper in detecting it's use.

At Computer Network Defence Ltd (CND) we will often test the resourcefulness of our new cyber security candidates by asking them to research a topic and deliver a short presentation on it.  The subject varies, however, one popular choice is "Phineas Fisher and The Cayman Bank". Given the technical nature of our candidates we ran the topic through ChatGPT to see what emerged and how good it is. The output is below:

Phineas Fisher is the pseudonym of an individual or group of individuals who have claimed responsibility for a number of high-profile cyber attacks and data breaches, including the hack of the Cayman National Bank and Trust in 2018. The attackers claim to have stolen millions of dollars from the bank and leaked sensitive information about its clients online. The identity of Phineas Fisher is unknown and the individual or group behind the attacks have not been caught by the authorities.

ChatGPT

Considering the response is generated by artificial intelligence, it's pretty impressive, what you can do is request ChatGPT to regenerate the response with more detail, as below:

Phineas Fisher is the pseudonym of an individual or group of individuals who have claimed responsibility for a number of high-profile cyber attacks and data breaches. They are most known for the hack of the Cayman National Bank and Trust in 2018. The attackers claimed to have stolen millions of dollars from the bank and leaked sensitive information about its clients online. The leaked data included information on wealthy individuals and companies, including details on their accounts, transactions, and personal information.

The identity of Phineas Fisher is unknown, and the individual or group behind the attacks have not been caught by the authorities. The group claims to be motivated by a desire to expose corruption and injustice, and has targeted financial institutions, government agencies, and other organizations that they believe are complicit in these issues.

The group is also known for publishing guides and tools for other hackers, encouraging them to target similar organizations and to use the same tactics and techniques that they have employed.

The group is considered as a hacktivist group, as they hack institutions for political or social purposes. It is important to note that, as of my knowledge cutoff, Phineas Fisher is still unknown, and the group's activities are still ongoing.

ChatGPT - More Detailed

There are so many other parameters to create content, for instance "500 words" and an essay style response will be produced within 500 words, it's remarkable.

Detecting the use of ChatGPT

I started by going to the horse's mouth and asking ChatGPT "How do I detect the use of ChatGPT" and got a response:

One way to detect the use of a language model like ChatGPT is to look for patterns or inconsistencies in the text that it generates. For example, you may notice that the text is too coherent or lacks the errors and inconsistencies that would be present in text written by a human. Additionally, you may be able to compare the text to other examples of text written by the same person or group to see if there are any significant differences.

Another way is to use GPT-2 Detection API like Grover (OpenAI) which can tell you if the text you are analyzing is likely to have been generated by a language model like GPT-2.

ChatGPT

The more obvious ways to suspect the use of ChatGPT would be to 

1. Run the question through ChatGPT yourself and familiarise yourself with the output, you'll find some repeated sentences. More advanced plagiarists with a lot at stake, such as submitting a thesis, may run their ChatGPT output through another AI tool which paraphrases the ChatGPT output.

2. Ask questions of the author to ensure they understand the output.

3. More onerous for you and the individual is to supervise the writing of the content. (unlikely)

4. Being cyber security specialists, we could block access to the ChatGPT website using their official computers, however, this is defeated using personal devices from home.  Instead, perhaps just have the technology report users and not block them, enabling you to be more diligent with the individuals of concern, who think they have got away with it.

Using Grover to Identify AI Generated Content

The proliferation of fake news is often driven by artificial intelligence and in order to help defeat this threat, tools have been created to read content. Grover being one of them, I ran the above ChatGPT content through Grover and received this response.

We are quite sure this was written by a machine


Report Print
×
Stay Informed

When you subscribe to the blog, we will send you an e-mail when there are new updates on the site so you wouldn't miss them.

 

Comments

No comments made yet. Be the first to submit a comment
Already Registered? Login Here
Thursday, 28 March 2024

Captcha Image

By accepting you will be accessing a service provided by a third-party external to https://www.cndltd.com/