Report | ChatGPT & Crime

What Law Enforcement Needs to Know about Large Language ModelsChatGPT and Crime Table of Contents Large Language Models and ChatGPT2 Intro1 Leveraging LLMs for Crime3 The Impact of Future LLM Developments5 The Benefits for Law Enforcement4 About Cognyte6

Introduction Although AI technologies have been around for several decades, the advent of ChatGPT has demonstrated in a tangible way the truly immense possibilities for AI-driven human-machine collaboration – both for legitimate purposes and for illicit aims. Criminals are beginning to utilize AI bots powered by Large Language Model (LLMs), such as ChatGPT, to accelerate their criminal activities. This eBook will explore the resulting challenges created for law enforcement, as well as how investigators and analysts can leverage LLM technology to stay one step ahead. Since being released to the public on Nov. 30, 2022, ChatGPT has been adopted at a stunningly fast pace, with 206.7 million unique visitors generating 1.76 billion visits during April 20231. While it took TikTok nine months following its global launch to reach 100 million users, ChatGPT crossed this milestone within two short months2. This incredible growth rate represents the fastest ramp- up in history of a consumer internet app, other than Meta's Threads app which leveraged Instagram's existing user base, and further reinforces the importance of law enforcement closely monitoring this phenomenon. Source: UBS 2 CHATGPT TIKTOK INSTAGRAM PINTREST SPOTIFY TELEGRAM UBER GOOGLE TRANSLATE9 30 41 55 61 70 78Time To Reach 100 Million Monthly Users* (excluding Threads) Large Language Models and ChatGPT2 Intro1 Leveraging LLMs for Crime3 The Benefits for Law Enforcement4 The Impact of Future LLM5 About Cognyte6 Intro1

Large Language Models and ChatGPT Large language models (LLM) are AI models which utilize deep learning techniques designed to understand and generate human language. LLMs are typically trained on vast amounts of diverse texts, such as books, articles, websites, and other sources, to learn statistical patterns, relationships, and language structure within the data. LLMs excel at a variety of natural language processing (NLP) tasks, including text completion, sentiment analysis, language translation, and question answering. They leverage generative AI techniques to generate coherent and contextually relevant text, mimicking human language patterns. ChatGPT is powered by the large language model GPT-3, which was developed by the AI research laboratory OpenAI. Its capabilities include: OpenAI offers an API enabling other companies and groups to use GPT-3 (as well as GPT-4, released in March 2023) to build their own AI chatbots and virtual agents. ChatGPT is currently the most well-known and commonly used AI bot, however there are many others, including Microsoft Bing and Google Bard, and many other LLMs, such as BERT, PaLM 2 and LaMDA, which are being actively used.Answering questions and providing explanations Explaining, writing, and improving code in programming languages including Python, Java, C++, JavaScript, PHP, Ruby, HTML, CSS and SQL Summarizing text Generating text based on a given prompt Translating text Large Language Models and ChatGPT2 Intro1 Leveraging LLMs for Crime3 The Benefits for Law Enforcement4 The Impact of Future LLM5 About Cognyte6 Large Language Models and ChatGPT2

Where LLMs fit into the AI domain Source: byteant, Cognyte Artificial Intelligence- Field of computer science which enables machines to imitate human intelligence Machine Learning - Branch of AI which provides machines the ability to learn automatically Deep Learning - Subfield of machine learning based on artificial neural networks Generative AI - Capability of AI systems to generate new content based on learned patterns & knowledge Large Language Models - Advanced AI models that can understand & generate human-like text Large Language Models and ChatGPT2 Intro1 Leveraging LLMs for Crime3 The Benefits for Law Enforcement4 The Impact of Future LLM5 About Cognyte6 Large Language Models and ChatGPT2

The Dark Side of ChatGPT: Leveraging LLMs for Crime While most users who leverage ChatGPT and other LLMs do so for legitimate purposes, signs indicate that criminals and bad actors have already begun utilizing these technologies for illicit activities. ChatGPT includes several safety features designed to prevent malicious usage which are updated by OpenAI on an ongoing basis. Many of these safeguards, however, can be easily circumvented through prompt engineering. While all the information ChatGPT provides is freely available on the internet, its ability to provide easily understood information and instructions in natural language makes it possible for bad actors to gain knowledge faster, enabling them to branch out into new areas or learn more advanced techniques and methods. Even if criminals continue utilizing the same techniques and business models as in the past, ChatGPT can accelerate the speed, scale, and efficiency with which they commit these crimes. Large Language Models and ChatGPT2 Intro1 Leveraging LLMs for Crime3 The Benefits for Law Enforcement4 The Impact of Future LLM5 About Cognyte6 Leveraging LLMs for Crime3

Criminals can utilize LLMs in numerous ways to accelerate cybercrime activities, including:How Criminals Can Leverage LLMs for Cybercrime Fraud and scams LLMs can enable more believable and effective fraud, social engineering and impersonation to scam victims into providing personal information, credentials or payments.

LLMs offer the ability to create more believable and legitimate sounding phishing emails, for example. In addition to quickly generating text based on simple prompts, LLMs can utilize original documentation, such as emails from a legitimate payments provider, to compose content that mimics the genuine author’s writing style, tone, and commonly used words3. Criminals can leverage LLMs to run automated bots, which can communicate in a convincingly mimic the communication style of real individuals or customer service representatives, to aid scams relating to phishing emails or malicious websites. Customized malware Experts are already seeing the first instances of cybercriminals using ChatGPT to develop malware. Even more concerning, bad actors will no longer need extensive development expertise or coding skills. ChatGPT lowers the bar for would-be hackers to try their hand at building effective malware and makes it easier for those with basic skills to develop much more advanced programs3. ChatGPT makes writing code in less common languages simpler for novices and experts alike. Anti-malware software tools have a harder time detecting malware in these languages, making it more likely to slip through the cracks of that security layer. Large Language Models and ChatGPT2 Intro1 Leveraging LLMs for Crime3 The Benefits for Law Enforcement4 The Impact of Future LLM5 About Cognyte6 Leveraging LLMs for Crime3

The cybersecurity vendor Checkpoint recently demonstrated that ChatGPT can successfully conduct a full infection flow, from creating a convincing spear-phishing email to running a reverse shell-attack5. 51% of IT decision makers believe there will be a successful cyberattack credited to ChatGPT within the coming 12 months (source: Blackberry) Criminals can utilize LLMs in numerous ways to accelerate cybercrime activities, including: Financial crime LLMs could be leveraged to effortlessly create false documentation on a mass scale to facilitate financial crimes. This could include generating fake supporting documents, customer profiles, transaction records, financial statements, invoices and contracts6. Human trafficking Traffickers often use social media platforms and messaging apps to recruit, groom, and exploit their victims. LLMs could be exploited to generate more effective text and content, and to create it faster, which would be leveraged to reach more victims7. Disinformation The proliferation of fake news and disinformation pose an increasing risk to national stability and public safety. LLMs have the potential to amplify this threat by enabling faster and more widespread generation of fabricated news articles, false or misleading information, and inflammatory content. Experts have only begun to explore how criminals could exploit LLMs, therefore law enforcement agencies must stay alert as additional use cases emerge.Leveraging LLMs in Other Criminal Domains Large Language Models and ChatGPT2 Intro1 Leveraging LLMs for Crime3 The Benefits for Law Enforcement4 The Impact of Future LLM5 About Cognyte6 Leveraging LLMs for Crime3

The Benefits of LLM Technology for Law Enforcement Leveraging LLMs for CTI – real world example LUMINAR leverages a large language model to provide a contextualized and personalized daily intelligence feed to cyber analysts by analyzing posts from diverse cyber threat sources, including deep and dark web forums, marketplaces and chat rooms. For each analyzed post, the system automatically:

• Extracts and generates new cyber-related entities, including threat actor, country of origin, name of campaign, and more
• Creates relations between entities, such as the threat actor, country of origin and targeted entities
• Summarizes the content of the post
• Extracts TTPs (Tactics, Techniques and Procedures) according to the MITRE ATT&CK framework This capability provides valuable and timely intelligence insights with targeted threat data relevant to each security organization.Although criminals and bad actors can exploit LLMs to scale up their activities, law enforcement authorities can also derive advantages from these technologies. Incorporating LLM technology into the software solutions utilized by authorities can assist investigators and analysts in performing their work faster and more effectively. Leveraging an LLM for LUMINAR’s intelligence feed enables far stronger capabilities than traditional text analytics. The analysis extends beyond the text contained in the posts being analyzed, as the model leverages its vast knowledge base to create new entities and establish connections to content that may not be explicitly stated in the analyzed text. For example, when analyzing a post about phishing or ransomware, the model can infer the specific attack technique used and tag it with the MITRE ID. In addition, in the case of texts or posts which contain slang, the LLM can leverage its knowledge base to understand those terms and parse the meaning in a cyber threat intelligence context.Cognyte’s LUMINAR is a Cyber Threat Intelligence solution (CTI) that enables SOC and intelligence teams to more accurately identify external threats, prevent attacks, and ensure resilience. Large Language Models and ChatGPT2 Intro1 Leveraging LLMs for Crime3 The Benefits for Law Enforcement4 The Impact of Future LLM5 About Cognyte6 The Benefits for Law Enforcement4

LUMINAR LLM-powered CTI intelligence feed Accelerating Investigations and Intelligence Analysis with LLMs LLMs have the potential to add significant value within many systems that support law enforcement investigations and intelligence analysis, including network intelligence solutions, decision intelligence platforms and more. For example, an AI-powered bot embedded within these systems could take natural language prompts or questions from the user, transform them into queries, filter the resulting data and present it to the user. The capability to ask questions in natural language would relieve analysts and investigators from the burden of creating complex queries. By utilizing LLM technology, the system can interpret terms and concepts that are not explicitly defined, such as recognizing that 'night-time' refers to the period between 8pm and 6am. Additionally, it can grasp contextual information from the preceding questions, enabling a deeper understanding of the user’s objective. Moreover, the AI assistant can suggest follow up questions and actions after the initial prompt, based on historical behavior and actions of system users, which the analyst or investigator may not have thought of on their own. Large Language Models and ChatGPT2 Intro1 Leveraging LLMs for Crime3 The Benefits for Law Enforcement4 The Impact of Future LLM5 About Cognyte6 The Benefits for Law Enforcement4

The Impact of Future LLM Developments Recent significant breakthroughs in the development of LLMs can be attributed to two key factors:

• Rapid advancements in supercomputers and deep learning algorithms
• Availability of massive datasets has allowed researchers to train their models on the vast input of information needed With LLMs now widely available and accessible, there are multiple avenues in which this technology could be leveraged - that could increase the threat posed by criminals and the resulting impact on law enforcement. The integration of LLMs with other AI technologies could make them even more formidable tools. For example, LLM- powered chatbots could be combined with generative AI technologies that can generate synthetic media, such as highly convincing deepfakes. This would make it even easier for criminals to conduct phishing, commit fraud, and spread disinformation. New AI bots are leveraging LLMs in a more autonomous way and thus could help criminals streamline their illicit activities and scale them up even faster. Auto-GPT, which uses OpenAI’s GPT-4 model, can take a broad goal inputted by a user and break it down into sub-tasks and execute each one – for example by reading documents, searching the internet, creating spreadsheets, and even downloading software – to produce the desired output without any further human input. Auto-GPT is designed to problem-solve, and thus handle more complex and multi-step projects autonomously.Another possibility is the emergence of “dark LLMs”, which could be hosted on the dark web to provide a chatbot without any ethical safeguards6. As AI technology advances, it is crucial for law enforcement to remain at the cutting edge of these developments. Taking a proactive approach will allow them to anticipate and prevent misuse by criminals while leveraging the potential benefits that these advancements offer. Large Language Models and ChatGPT2 Intro1 Leveraging LLMs for Crime3 The Benefits for Law Enforcement4 The Impact of Future LLM5 About Cognyte6 The Impact of Future LLM5

To learn more Footnotes:

1. Source: SimilarWeb
2. Source: UBS, SensorTower
3. Source: Enzoic
4. Source: Checkpoint, January 2023
5. Source: Checkpoint, December 2022
6. Source: sanctions.io, March 2023
7. Source: TS2, April 2023
8. Source: Europol Tech Watch Flash – ChatGPT, March 2023Cognyte is the global leader in investigative analytics software that empowers a variety of government and other organizations with Actionable Intelligence for a Safer World™. Use of these products or certain features may be subject to applicable legal regulation. The user should familiarize itself with any applicable restrictions before use. These products are intended only for lawful uses by legally authorized users. Not all features may be available in all jurisdictions and not all functionalities may be available in all configurations. Unauthorized use, duplication, or modification of this document in whole or in part without the prior written consent of Cognyte Software Ltd. is strictly prohibited. By providing this document, Cognyte Software Ltd. is not making any representations regarding the correctness or completeness of its contents and reserves the right to alter this document at any time without notice. Features listed in this document are subject to change. Contact your Cognyte representative for current product features and specifications. All marks referenced herein with the ® or TM symbol are registered trademarks or trademarks of Cognyte Software Ltd. or its subsidiaries. All other marks are trademarks of their respective owners. © 2023 Cognyte Software Ltd. All rights reserved worldwide.About CognyteDiscover the top tech trends impacting law enforcement >>Report Contact us Learn more about leveraging AI for law enforcement >> Large Language Models and ChatGPT2 Intro1 Leveraging LLMs for Crime3 The Benefits for Law Enforcement4 The Impact of Future LLM5 About Cognyte6 About Cognyte6