Generative AI is a rapidly evolving field and at its core, generative AI is about teaching machines to create something new, whether it's a piece of music, a story, or even an image. One of the most exciting applications of generative AI is in the field of natural language processing (NLP), where advanced language models like GPT (Generative Pre-trained Transformer) can generate human-like text based on a given prompt. This has huge implications for open-source intelligence (OSINT), where the ability to generate text quickly and accurately can help investigators uncover hidden information and insights.
In this blog post, we'll explore how you can use ChatGPT to enhance your OSINT investigations and analysis workflows. We'll cover tips for effective prompts, the risks and challenges of using generative AI in OSINT, and the benefits of integrating ChatGPT into your toolkit.
What are the benefits for OSINT?
By leveraging the power of generative AI, investigators and analysts can gain new insights, save time, and improve the accuracy and efficiency of their OSINT workflows. Here are some examples of how it can be used:
Automate repetitive tasks, freeing up more time for higher-level analysis.
Identify patterns and anomalies in large datasets that may be missed by human analysts.
Improve communication and collaboration among investigators by generating summaries or reports based on their findings.
Process large amounts of content quickly and break down complex problems into smaller, more manageable pieces.
Learn and improve accuracy over time through unsupervised and self-supervised learning.
There are many workflow examples, in which we have broken down into three areas: (1) Research assistance (2) Efficiencies (3) Technical Enablement
How does it differ from "Google Searching"
While Google search returns results based on keywords and directed filters (such as time, domains, and attributes), generative AI systems like ChatGPT take a different approach. Instead of providing a list of relevant content, ChatGPT can synthesize information from large datasets into a human-like response. We look at it as another tool in the toolbox, to potentially uncover new insights and connections that may not be immediately apparent through traditional search methods.
Prompting Tips For Better OSINT Workflow
When using ChatGPT for OSINT workflows, it's important to use specific prompts that will help generate relevant and accurate responses. It's also essential to understand the limitations of ChatGPT, such as its inability to verify the accuracy of the information it generates. As such, it's recommended to use ChatGPT-generated responses as a starting point for further investigation and analysis, as well as for generating new ideas. Here are some pointers for effective prompting:
Ask questions in a human-like, conversational manner.
Give ChatGPT context and constraints to work with.
Use prose (sequence) to chain together related questions and responses, also called chain-of-thought sequencing.
Practice, and refine your prompts over time.
Let's run through some prompting examples:
In one case, we want to develop further context for an emerging situation:
Find news articles of interest from reputable sources.
Copy the body of text for processing into ChatGPT.
Ask ChatGPT to "extract names, locations, and organizations into a table, with a short description of each, from this body of text".
Ask ChatGPT to "provide detailed background information on the entities (be specific) mentioned above, including historical violent actions and their capabilities".
Ask ChatGPT to "create an issue brief template report and populate it with information on the threat groups found above, including source references, and expand on the research with additional information about the threat groups involved, including major conflict areas within the country and a table list of the major leaders of each group."
It's important to note that while ChatGPT can provide a new and natural workflow for researching ideas and expanding on information quickly, validation remains paramount. We should always verify the information we obtain through other sources before acting on it.
Another example of how ChatGPT can assist in OSINT is with travel support. For instance, if our task is to prepare an overseas travel brief for someone traveling to Country Z and staying in City X, our workflow might look like this:
Ask ChatGPT to "generate a travel brief for the major threats in Z and create a table of the nearest hospitals, police stations, and military bases in X. Be sure to include ground and air transportation options."
Validate the information with sources such as the CIA Factbook, FVEY government travel advisory sites, and other open-source reporting.
Alternatively, you may want to have this information in a table format. In this case, we would use the prompt: "Create a list of all hospitals, fire stations and police stations within a 2km radius of the Melbourne Cricket Ground. Title column one as type, with the options of Police Station, Fire Station and Hospital. Title column two as Name, and list the names of the venues underneath. Title column three as Location and list the address underneath."
Overall, these examples demonstrate how ChatGPT can streamline our OSINT workflows and enable us to generate new ideas and insights quickly. However, we must remain diligent in verifying the accuracy of the information we obtain through ChatGPT, as well as any other sources we use.
Risks and Possible Issues of Using ChatGPT for OSINT
There are several risks and considerations to keep in mind when using ChatGPT for OSINT workflows. These include:
Operational Security: Sharing sensitive information or conducting investigations using ChatGPT may compromise operational security. As such, it's important to ensure that any information shared with ChatGPT is not sensitive or confidential.
Web platform vs API: Using ChatGPT through a web platform may not provide the same level of security and customisation as using the API. For example, its inclusion in NexusXplore can help to identify key items on information and rapidly pivot into other collection tools within the platform.
Privacy, data retention and storage: ChatGPT does store data and responses generated during its use for at least 30 days across all versions (at time of writing) which could pose a privacy risk. ChatGPT may generate responses that are subject to international privacy laws, such as the GDPR.
Accuracy and sourcing: ChatGPT's responses may not always be accurate, and it's essential to verify any information generated using other sources, and assessing the information for relevance, reliability, credibility, and corroborate the information (or R2C2 for short).
Copyright: ChatGPT may generate responses that infringe on copyright laws. Additionally, at this time we don’t know what data sources were used to generate the model.
Ethical use: Using ChatGPT for malicious purposes or in violation of ethical principles could have significant consequences. Unfortunately, ChatGPT can be manipulated to generate responses that are harmful or misleading.
Introducing the OSINT Research Assistant (ORA) for NexusXplore
NexusXplore is the world-leading platform used for open-source intelligence, and offering unparalleled capabilities to help analysts collect and analyse data. The team have recently integrated ChatGPT into NexusXplore, so that analysts can enhance their workflows for even more efficient and comprehensive analysis processes.
The integration of ChatGPT into NexusXplore's ORA offers several key benefits on top of its use within the browser. These include:
No requirement to sign up using a mobile phone and email, reducing attribution considerations for government and corporate alike.
Queries entered into ORA are all stored as separate conversations. This means that queries won't pool within the Open AI servers and therefore won't provide an indication as to the broad industry, background, or intended purpose of the user.
The ability to rapidly pivot from other data collection in the NexusXplore platform to ORA, allowing for quick and efficient analysis of large volumes of data, to quickly derive meaning and insights.
Analyst can identify key items on information from the ORA response and rapidly pivot into one of the many dedicated tools of NexusXplore, for example, gaining further company and supply chain insights, or enhancing situational awareness through geotagged social media and mobile device mapping.
The ORA's language translation capability allows analysts to leverage AI to gain insights from sources in multiple languages.
With Great Power Comes Great Responsibility
ChatGPT is a powerful tool that can be used to enhance OSINT workflows within and out of NexusXplore. However, it's essential to understand the risks and considerations associated with its use and take steps to ensure that it is used in an ethical and responsible manner and understanding the importance of verifying information online.
To delve deeper into the learnings above or discuss how we can more broadly support or assist in uplifting your organisation's OSINT capability, please contact us about our NexusXplore platform, or our in-person and self paced training courses.