AI in Intelligence Analysis

By Dr. David James Kritz  |  10/20/2026

The U.S. Intelligence Community is composed of a civilian and military workforce. This workforce follows an intelligence cycle that takes raw data or information, transforms that data and information into intelligence, and distributes that knowledge to its customers in the form of a product. The product addresses customers’ questions and is also known as priority intelligence requirements.

The intelligence cycle is comprised of several sectors: planning, collection, processing, analysis, dissemination, and evaluation. Recently, artificial intelligence (AI) has been incorporated into the analysis sector of the intelligence cycle to aid analysts.

The Hierarchy of Communication in The Intelligence Community

Artificial intelligence in intelligence analysis is a new phenomenon that can be considered divisive, depending on who is asked and how AI is used.

In the U.S. Intelligence Community, information is transformed into intelligence once analytic tradecraft is applied. Until recently, intelligence analysis was conducted solely by humans. Now, artificial intelligence should be considered a tool or as a partner to help collect and refine information.

Unfortunately, intelligence analysts having to wade through a large amount of data is not a new phenomenon. As the volume of open-source information added to the internet grows daily, open-source collection is growing as well.

Depending on the artificial intelligence models, large language models (LLMs) can be beneficial to intelligence analysis. They can automate data collection from multiple sources such as:

• Articles
• Books
• Documents
• Massive datasets
• Reports
• Patents
• Transcripts

In addition, AI tools can analyze information much faster than humans as well as make sense of previously misunderstood criteria, answer questions, and find patterns. AI can also aid analysts in drafting intelligence documents for decision makers.

As a result, AI models can help produce actionable insights and perform predictive analyses. Predictive analyses use previous information and can help to forecast future events based on probability.

Variables That Affect AI in Intelligence Analysis

There are two key components that affect the reliability and validity of the output from AI software, which are the type of AI model and the quality of the data. It takes a team of people – including data scientists led by organizational leadership – to help ensure the correct AI model is applied to intelligence analysis.

Remember the saying, “Garbage in, garbage out?” This saying certainly rings true when AI is used to create prompts and determine the quality of the intelligence product.

Intelligence analysts are responsible for providing quality prompts and context. That will allow AI tools to create meaningful and relevant responses for the analysis of strategic, operational, and tactical national intelligence topics.

Natural Language Processing and Its Role in Intelligence Analysis

A branch of artificial intelligence is natural language processing. The natural language processing of large language models allows artificial intelligence software to read text, translate text into different languages, and perhaps most importantly, understand that text’s meaning.

The use of human intelligence and signals intelligence is paramount not only for translating foreign languages, but also for distinguishing between people with different dialects or who use informal language such as slang. Intelligence analysts can then establish a deeper understanding of information for customers, especially when words have multiple meanings.

Faster natural language processing in AI software can help ground forces to gain greater situational awareness and work through complex messages. AI could quickly determine if the enemy is talking about a piece of fruit or using a code word for an improvised explosive device. As a result, the military would be better positioned to achieve mission effectiveness toward national interests in an operational environment.

Red Teaming in Intelligence Analysis

Challenging the status quo is rarely an easy endeavor. When it comes to using AI to conduct intelligence analysis, there may be some resistance to the suggestion that entrenched paradigms may no longer be valid or should remain as preferred practices. Consequently, analytic tradecraft may need to evolve.

For instance, intelligence staff could adopt AI software for red team analysis. According to The Mind Collection, “Red team analysis, also known as red teaming, is an imaginative analytical technique that attempts to replicate the thoughts and actions of an adversary.”

By incorporating red teaming to challenge the views expressed by analysts, red teaming causes analytical refinement to produce other intelligence assessments. This way of using red teaming could be beneficial prior to senior leaders creating official policy.

Creating Trustworthy Outputs When Using AI in Intelligence Analysis

Artificial intelligence is not without its challenges. For instance, one challenge is creating a trustworthy output with AI tools. Current AI software can make errors, which are known as hallucinations. Not only is the output incorrect, but in some cases AI’s responses to prompts are completely fabricated and presented with certainty.

Will AI hold up to the scrutiny of senior government leaders when they ask questions about intelligence assessments prior to issuing an executive order? Will it be acceptable for an intelligence analyst who provides a briefing to respond with a statement like, “Well, ma’am/sir, this output is what the algorithm produced”? That’s unlikely to happen.

It is vital to ensure that the correct AI tools are selected for use in national intelligence. Applying AI to government work focused on national interests is relatively new.

Another way to think about analysts using AI is that it is similar to researchers using the scientific method when conducting research. The speed at which AI answers questions is amazing but it is also a reason why analysts need to be vigilant and rigorously check the output.

One potential way to work through AI tech issues and help AI tools to simultaneously improve performance and build analytic skills with AI is to feed training data into foundation models. Then, intelligence analysts could verify its validity.

Security Risks of Using AI for Intelligence Analysis

Another challenge with using AI in intelligence analysis is security. Intelligence analysts primarily work with classified information, so it’s important to be mindful that AI models will be used in sensitive, compartmented information facilities. Consequently, there may be some classification and integration issues associated with developing, incorporating, and maintaining AI technology.

I am a retired intelligence officer. As AI gained more attention for its potential as a disruptive technology, I often wondered how long it would take, if ever, for intelligence analysts to discover whether an adversary had hacked into the system and planted disinformation in the algorithms. Numerous new security risks related to national defense may occur with AI adoption and integration.

Imagine what would happen if global conflict occurred over disinformation from AI tools. Here are six different types of attacks that could adversely affect the existing AI tools used in intelligence analysis, create a mistrust of AI, and negatively affect decision making:

• Adversarial inputs – Deliberate misinformation could be planted into a system for malicious purposes and cause an AI tool to produce false or misleading information.
• Data poisoning – Poisoned training data could deliberately mislead an AI model's learning and data collection. As military intelligence focuses on tactical, operational, and strategic areas, data poisoning can be considered the most dangerous as it centers on the strategic level and affects training data. It also allows for backdoor attacks so that when a trigger occurs, the AI model produces material that benefits the enemy.
• Manipulation of AI agents – AI tools such as ChatGPT®, Claude®, or Gemini® can access external information and interact with other systems. This situation is concerning because AI continues to become more autonomous.
• Model inversion attacks – By reverse-engineering the original AI model, model inversion attacks focus on extracting personal, privacy-related information such as medical records, patient information, and Social Security numbers.
• Model jailbreaks – These attacks primarily focus on large language models and can manipulate AI systems and AI chatbots by finding system vulnerabilities. Because AI chatbots are designed to be helpful, some hackers take advantage of AI chatbots. This type of attack is also unique because it can focus on roleplaying and personas that may be harmful to the recipient interacting with the chatbot.
• Prompt injection attacks – With this type of attack, an attacker deliberately writes prompts that tell the AI to ignore previous instructions, thereby creating malicious acts. Similar to model jailbreaks, this type of attack focuses on attacking large language models. Prompt injection attacks bypass safety standards and inject malicious instructions. According to the Open Worldwide Application Security Project® (OWASP®) Foundation, malicious prompts can be embedded into images, audio files, or video files that large language models scan for information.

Maintaining Security for AI in Intelligence Analysis

Accuracy and timeliness are two vital criteria that are required by intelligence customers. Because AI tools receive access to classified information, trustworthiness and the mitigation of data leakage are other concerns.

Imposing additional security controls will likely affect the dissemination time for intelligence products and perhaps customer confidence. The stronger the security controls, the less useful AI may be for analysts. Senior organizational leaders and security managers will need to create a risk management balance between access and security concerns.

The Future of AI in Intelligence Analysis

Using AI in intelligence analysis has both its advantages and disadvantages. Human intelligence analysts will still be needed to verify information accuracy and validity, as well as to provide context for the AI tool’s output.

AI software can provide predictive analysis, speed, and pattern identification. It can also identify relationships, produce insights, and offer multiple potential solutions to a problem. AI’s ability to create an optimal solution for homeland security and other national interests outweigh its limitations, especially since artificial intelligence’s technology and capabilities will likely improve over time.

It’s essential to remain mindful that, like the other intelligence disciplines, AI and machine learning should be used as a support aid and outputs will still require close scrutiny and human judgment. It’s equally important to not cross a line and use AI to outsource human thinking in the decision-making process.

The Bachelor of Arts in Intelligence Studies at APU

For students who want to improve their knowledge of the U.S. Intelligence Community, intelligence collection, and other related topics, American Public University part of American Public University System (“APU”) offers an online Bachelor of Arts in Intelligence Studies. For this degree program, students can enroll in courses such as international relations, law and ethics in intelligence studies, and threat analysis. Other courses include foreign intelligence organizations and research design and methods.

This bachelor’s in intelligence studies also features multiple concentrations to enable students to take courses suited to their professional interests:

• General
• Asian Area Studies
• Counterintelligence
• Criminal Intelligence
• Cyber
• Geospatial Intelligence
• Intelligence Analysis
• Intelligence Collection
• Intelligence Operations
• Latin American Area Studies
• Middle East Area Studies
• Terrorism Studies

For more information, visit APU’s security and global studies degree program page.

ChatGPT® is a registered trademark of Open AI OpCo, LLC.

Claude® is a registered trademark of Anthropic, PBC.

Gemini® is a registered trademark of Google, LLC.

Open Worldwide Application Security Project® and OWASP® are registered trademarks of The OWASP Foundation, Inc.