For mankind, technology is an effective way to build a better future. For a long time, humans have dreamt of powerful technological advancements such as nuclear fusion, artificial intelligence (AI), and quantum computing. Such technologies mark a huge leap in advancing as a civilization and were widely considered to be mere products of imagination only seen in movies and read about in books.
Although humans have been on the verge of cracking these technologies for a while now, AI-based technology has not yet been optimized to be deployed as commercial products. This is why it's easy to understand the hype when a company announces that it is close to commercializing one of these technologies, case in point, ChatGPT, which operates under the technological umbrella of AI.
ChatGPT was developed by a San Francisco-based company called OpenAI that specializes in AI research and development. It was founded in 2015 with Elon Musk as one of its co-founders. Based on the Generative Pre-trained Transformer 3.5 (GPT-3.5) language model, ChatGPT is a large language model (LLM) that is optimized for human-conversation, meaning it can clarify on questions posed to it by asking follow-up questions, identify its own errors, and debate on topics much like a human.
The technology that ChatGPT is based on (GPT-3.5: an improved version of its successor, GPT-3.0), is the next generation of OpenAI's GPT series and is considerably more powerful than its predecessors. OpenAI stated that the GPT series was created for enhancing language generation models by fine-tuning them for particular tasks after training them on broad sets of data.
Although the GPT series has been around for a while, the latest iteration crossed the standard threshold of AI tools. With more than 175 billion machine learning parameters (meaning that it can generate comprehensive, coherent, and appropriate content on a large spectrum of topics with minimal informational input), GTP-3 is one of the largest machine learning neural network models ever created. GPT-3 was integrated with chatbots in order to create a variant endowed with human-like conversational and content-generation abilities, which was named ChatGPT.
ChatGPT is very different from other commercial chatbots. While other chatbots utilize the power of the internet to gain access to information and form responses according to its findings, ChatGPT is not connected to the internet, and therefore does not have access to information available online. It being disconnected from the internet is not unintentional, as GPT-3 was not designed to scour the internet for information but to primarily comprehend and mimic human-like content; ChatGPT was not designed to be a search engine, but a language processing system.
All the content formulated by Chat-GPT is based on data that it has been fed from books, websites, articles, and other various sources. This means that it has been pretrained to generate responses by interrelating specific words with specific concepts without any external data consumption via the internet.
The spectrum of the data it has been trained on is significantly vast, and it is then calibrated to carry out particular tasks like writing a synopsis, translating text, or even formulating legal documents. This makes it highly efficient, as the content produced is context-specific and holds enough weight for it to be reasonable and relevant to the conversation and the user.
In short, OpenAI's ChatGPT can:
The significance of ChatGPT cannot be undervalued, especially in the cybersecurity industry. It uses the revolutionary technology of AI, which has been sought after by humans for a long time. Although AI has been in use for a long time by companies like Google and Microsoft, OpenAI enabled and allowed public-experimentation of ChatGPT by creating a user interface for it.
Casey John Ellis, founder, chair, and chief technology officer of an IT security company called Bugcrowd, stated "It's frankly influenced the way I've been thinking about the role of machine learning and AI in innovations."
Other cybersecurity professionals landed on a similar conclusion. Researchers who spent a considerable amount of time testing ChatGPT found that if not AI as it is, machine learning has the potential to be a revolutionary and disruptive technology. The researchers performed several defensive cybersecurity tasks and offensive cybersecurity campaigns like drafting polished or persuasive phishing emails, developing functional YARA rules (malware identification methods), and creating evasion codes that can help attackers bypass threat detection. And these are just the tip of the iceberg when it comes to the implications of AI and ChatGPT in cybercrime and cybersecurity.
While there are some cybersecurity professionals who aren't as surprised by its capabilities and applicability, most have expressed concerns on how cybercrime is going to change. For example, Steve Povolny, principal engineer and director of cybersecurity firm, Trellix, said, “While ChatGPT attempts to limit malicious input and output, the reality is that cyber criminals are already looking at unique ways to leverage the tool for nefarious purposes. It isn’t hard to create hyper realistic phishing emails or exploit code, for example, simply by changing the user input or slightly adapting the output generated.”.
Both the positive and negative effects of AI have already started manifesting across all industries. On the one hand, the premise for applicable and commercial AI-based technology has been set, while on the other hand, cybercriminals have discovered how to convert the technology and leverage ChatGPT to carry out intricate and targeted cyberattacks.
We have covered what ChatGPT is, how it operates, and the technology behind it. The next part in our Exploring ChatGPT series will cover ChatGPT in relation to AI-driven cybercrime and cybercrime cases to learn from to get a better understanding of how cybercriminals manage to trick ChatGPT into helping them with cybercrime.