AI and the Future of National Security
How can national security organizations best leverage AI to improve their operations?
When thinking about the role of artificial intelligence (AI) / machine learning (ML) in national security, people often think of the fancy robots, killer drones, and supercomputers that grace our TV and movie screens. However, AI will do much more for the national security of the US and its allies than lead to dystopian nightmares. AI and ML will be instrumental in a wide range of tasks critical for national security, offering rapid, data-centric decision-making and optimization capabilities. The use of AI by national security organizations certainly isn’t new. As I’ve outlined in my previous work, the US and allied militaries have used AI and autonomy for a number of years. However, advancements in generative AI, data collection, and computing technologies have renewed attention and excitement about the potential for AI to improve national security.
Over the past few years, I’ve spent a great deal of time digging into the ways that AI can best be used to solve real problems in national security. I like to put the artificial intelligence technologies that will benefit the national security of the US and its allies (both government and commercial customers alike) into two primary buckets: “horizontal”, AI-enabling infrastructure and “vertical” applications of AI to relevant tasks.
Before the US government is able to build out any useful “vertical” AI-enabled applications, it’s critical for the US and its allies to build out robust, “horizontal” AI-enabling infrastructure. Infrastructure includes:
Foundational models built on comprehensive, secure training datasets
Technology that enables the development of cost-effective models for specific tasks
Technology that enables ML algorithms to run at the edge
Testing for ML algorithms
Monitoring and observability of ML algorithm performance
Security for ML algorithms against prompt injections, data poisoning, model drift, etc
Role based access control for ML algorithms and the data they rely on
Data tooling to prepare data for ML applications (ETL1, data labeling, data organization, etc)
Model repair and explainability
Secure hardware and compute technology
Much of this AI-enabling infrastructure is already used by digital-native commercial organizations. However, for large, legacy organizations like the US government (and particularly the Department of Defense and intelligence community), adoption of AI-enabling infrastructure will be crucial to AI adoption. To drive home the point, Craig Martell, the Director of the Chief Digital and AI Office (CDAO), the primary group within the Department of Defense (DOD) responsible for AI adoption, has repeatedly stressed that the DOD should not start building AI applications…yet.
First, the DOD needs to produce and organize its high quality data. The DOD is not an organization that employs a large number of AI experts. However, what the DOD lacks in AI expertise, it makes up for with its large amount of unique data – decades worth of design and operations manuals for their hardware systems, war games, geospatial intelligence, signals intelligence, human intelligence reports, cutting-edge scientific research, etc. Currently this data is siloed, unorganized, overclassified, and difficult to find. So, Martell has outlined the DOD’s plan for a “Data Mesh” – essentially a federated data catalog that organizes all of the DOD’s data and allows AI experts to search for that data in one place (one step along the pathway to creating a DOD-wide data lake). Of course, it’s not reasonable for the DOD to move all of its data into one centralized location, but it can (and should) catalog its data and store that catalog in one place so that the AI experts can find the data they need to develop powerful AI-enabled tools. Once the DOD has its data organized, it can outsource the development of AI applications to AI experts who have access to the DOD’s data catalog.
Already startups are working to help the DOD and other large legacy organizations with siloed data develop federated data catalogs to organize data in one place. For example, “Data Fabric” and “Data Mesh” companies like Nexla and Cinchy use metadata to help its customers organize and integrate their data for AI and analytics use cases without the need to centralize all of an organization’s data in one place.
A data mesh is just the tip of the iceberg for the kinds of AI-enabling infrastructure the DOD will make use of in the years to come. In addition to a data mesh to organize the DOD’s data, the DOD will also need data tools like data labeling and ETL that help prepare the data itself for ingestion by AI applications. The DOD and intelligence community (IC) are highly reliant on unstructured documents made up of natural language (human intelligence reports, after action reports, etc) that need to be turned into structured data that can be ingested by ML algorithms. The DOD and IC will need to make use of companies like Unstructured to prepare its unstructured data for use by AI applications.
Finally, once the DOD actually develops and deploys AI applications, it will need tools that enable it to test, track the performance of, explain, and secure its AI applications.
Once organizations have secure AI-enabling infrastructure in place, they can start to use AI in “vertical” AI-enabled applications that actually make use of their data. Tasks that benefit most from new advances in AI will be tasks that benefit from data-driven decision making and optimization. These include tasks like:
Cybersecurity: The use of ML in cybersecurity is by no means new. For years, cybersecurity companies have used ML to detect malicious behavior like phishing attacks, malware, and data exfiltration. New advances in generative AI will continue to push the needle forward in improving cybersecurity. Large language models (LLMs) will make it easier to ingest threat intelligence reports and turn them into detection rules. They will assist penetration testers in their work and will be able to reverse engineer binaries. Additionally, AI will generate playbooks during cyberattacks, predict attackers’ next move in the midst of cyberattacks, automate fuzzing2, speed up security reviews, assist with log analysis, and perhaps even help organizations transition code bases from unsafe languages like C/C++ to memory-safe languages like Go and Rust.
Autonomy-enabling technologies: The field of autonomous systems is no stranger to AI. Already, projects like DARPA’s AlphaDogFight have shown that (in simulation) AI is better than human fighter pilots at piloting fighter jets. Furthermore, new advances in generative AI and computing technology (especially edge AI) will accelerate the development and performance of autonomous systems. For instance, generative AI products developed by companies like Resim AI will help speed up autonomous systems development through improved simulation and testing and improve human-machine teaming by enabling humans to communicate with autonomous systems using natural language. Additionally, today, autonomous systems engineers rely on log analysis to troubleshoot their systems. Advances in AI-enabled log analysis will make it much easier for engineers to troubleshoot their systems. Further, synthetic data generation has the potential to improve computer vision, and advances in AI-enabled sensor fusion will improve the autonomous systems’ ability to perceive the world. Improvements in edge AI will allow autonomous systems to actually make use of advances in ML at the edge. Advancements in autonomous navigation will enable autonomous systems to operate in swarms without human intervention in RF-denied environments. Finally, there are a whole host of new AI-enabled design tools that will make it easier for engineers to design the hardware needed for their autonomous systems.
Military decision making: AI has the potential to significantly enhance military decision-making by providing advanced analytical capabilities and real-time intelligence. AI can process and analyze vast amounts of data from various sources, such as satellite imagery, drones, sensors, and intelligence reports, much faster and more accurately than human analysts. This capability enables the military to gain comprehensive situational awareness, identify potential threats, and make informed decisions swiftly. AI-powered predictive analytics can anticipate adversary actions, suggesting optimal strategies and tactics. It can also simulate various scenarios, helping military leaders to evaluate the potential outcomes and risks of different courses of action. Additionally, using AI in military decision making can increase the speed at which militaries can make decisions which is a key tactical advantage in kinetic warfare. Already organizations like Palantir, and Scale AI have released AI-enabled military analytics and decision-making platforms (Palantir AIP and Scale Donovan respectively).
Geospatial intelligence analysis: In recent years, as satellites have proliferated, the amount and diversity of geospatial data collected has exploded. In order to make sense of all the data now being collected from space, national security organizations will need to use advanced AI to organize and analyze this data. Technology that will enable effective geospatial intelligence analysis include: 1) edge AI algorithms that are able to process data on satellites in space in order to reduce the amount of data being sent down to earth in the first place, 2) analysis platforms that enable analysts to use natural language processing to search satellite data, and 3) improved computer vision algorithms that automate the extraction of meaningful insights from geospatial data sources. These capabilities paired with the proliferation of satellite data will enable real-time monitoring and analysis of terrain, environmental changes, urban development, and military movements. Moreover, AI's ability to integrate data from multiple sources and its advanced pattern recognition capabilities can uncover subtle and complex correlations that might elude human analysts, which will enhance predictive analytics and enable governments and organizations to anticipate and respond proactively to potential crises or opportunities.
OSINT analysis: Increasingly, the IC relies on open source intelligence analysis including commercial satellite data, social media, public records, and other unclassified public data. AI can enhance OSINT analysis by automating and refining the collection and analysis of publicly available data. AI can efficiently sift through vast quantities of data from diverse sources such as social media, news outlets, blogs, forums, and public records, extracting relevant information much faster than traditional manual methods. This capability is particularly useful in identifying trends, patterns, and connections that might be missed by human analysts. For instance, AI-driven natural language processing (NLP) tools can analyze text from multiple languages, understanding context, sentiment, and subtle nuances, which is crucial in intelligence gathering. AI can also assist in real-time monitoring and alerting, flagging potential security threats, disinformation, or important events as they unfold, thus enabling quicker responses. In the realm of image and video analysis, AI technologies like machine learning and computer vision can recognize faces, objects, and activities, providing valuable insights for security and intelligence purposes. For instance, companies like Vannevar Labs enable intelligence analysts to integrate and analyze OSINT, including text data in foreign languages.
Enterprise search / knowledge management: The US government is a tremendously large and complex organization with many siloed data sources. AI-enabled enterprise search will enable the DOD and IC to connect and search all of its siloed data sources efficiently and effectively. Of course, any enterprise search / knowledge management system used by national security organizations will need an understanding of classification levels in order to maintain data security.
Text and code generation: AI will improve the efficiency of national security organizations by enabling fast text and code generation. Just like commercial organizations, the DOD and IC rely on humans to write reports and memos. Tools that enable efficient summarization of information and generation of reports, memos, and code will greatly improve the efficiency of national security organizations. Additionally, improved code generation will enhance the efficiency of those building software and autonomous systems for national security use cases. For instance, some startups are developing specialized code generation software for tasks like sensor integration and secure robotics programming.
Manufacturing optimization: In recent years, a number of organizations have begun using AI to optimize manufacturing processes and hardware design. For example, Rangeview uses AI to improve the designs of their 3D printed molds in order to reduce defects and speed up processing. Additionally, a number of AI design “copilots” have emerged in order to help optimize hardware and electronics design to improve efficiency of manufacturing and performance. In addition to improving the designs of the items being manufactured, technology has also emerged to help organizations analyze IoT data collected from sensors in factories. By analyzing this data, organizations can improve the efficiency of their manufacturing processes. Further, AI can be used for predictive maintenance of machinery, reducing downtime and extending the lifespan of equipment by anticipating failures before they occur. AI-driven robots and automation technologies can perform tasks with greater accuracy and speed than human workers, particularly in hazardous or repetitive work environments, and AI can be used for quality control, detecting defects as they arise. AI-enabled manufacturing optimization will improve national security by increasing the effectiveness and efficiency of critical infrastructure, including the defense industrial base, and improving American manufacturing resilience.
Supply chain management / security: Organizations are paying increased attention to their supply chains as geopolitical winds shift and events like the COVID-19 pandemic showcased the fragility of global supply chains. A significant amount of supply chain data (shipping data, communications with vendors, hardware specs, etc) is in non-standardized, unstructured form, as are many rules and regulations surrounding supply chain management. Generative AI can help organizations interpret and analyze unstructured supply chain data to ensure resiliency and robustness in critical supply chains. Additionally AI will enable organizations to automate parts of their supply chain management and security and will help bring visibility to organizations’ supply chains.
These use cases are just scratching the surface, and each subject described here deserves its own blog post. Going forward, I plan to dive into each subject in a dedicated essay to fully explore how AI will improve each facet of national security.
Let me know your thoughts! I know this is a quickly changing technology space and welcome any and all feedback. Where else are there opportunities for advancements in AI and ML to revolutionize national security? And please do not hesitate to reach out if you or anyone you know is building AI-enabled products at the intersection of national security and commercial technologies!
Note: The opinions and views expressed in this article are solely my own and do not reflect the views, policies, or position of my employer or any other organization or individual with which I am affiliated.
ETL stands for “extract, transform, and load” and is used in data warehousing to consolidate an organizations’ data from multiple sources into a single, coherent framework. For more, see “AWS - What is ETL?”
Fuzz testing or fuzzing is an automated software testing method that injects invalid, malformed, or unexpected inputs into a system to reveal software defects and vulnerabilities. For more, see: “Fuzz Testing” from Synopsys