How AI Can Enhance The Intelligence Community's Ability To Provide Strategic Warning
Hello, I’m Ylli Bajraktari, CEO of the Special Competitive Studies Project. In this edition of our newsletter, SCSP’s Nandita Balakrishnan discusses the IC’s urgent need to deploy AI tools as a means to improve the speed and accuracy of strategic warnings to policymakers. For more analysis on the use of AI for strategic warning, including on the current technological state-of-play, see SCSP’s joint expert analysis with the Alan Turing Institute’s Centre for Emerging Technology and Security.
Genesis: Artificial Intelligence, Hope, and the Human Spirit
SCSP Chair Dr. Eric Schmidt continues the pursuit of groundbreaking insights in his new book released today, Genesis: Artificial Intelligence, Hope, and the Human Spirit. This work, co-authored with Dr. Craig Mundie and the late Dr. Henry Kissinger who encouraged us to establish SCSP, explores the complex relationship between humanity and AI, offering a unique perspective shaped by Kissinger's profound historical understanding. Delve into the pages of "Genesis" and discover a thought-provoking exploration of AI's impact on our world.
SCSP x AGI House x George Mason University Hackathon
Last week, we hosted a thrilling hackathon focused on cutting-edge AI and national security challenges. With nearly 100 attendees, including experts from the DOE, CIA, NSC, DIU, and CISA, we witnessed incredible innovation and collaboration. Over 80 hackers formed 20+ teams to tackle real-world problems. Congratulations to our winners - you can view all of the presentations here!
1st Place: Spectre Red Teaming - Agentic AI approach to jailbreaking LLMs
2nd Place: SentinelPilot - Intelligent Fault Detection and Recovery for Safeguarding Critical Infrastructure
3rd Place: EarthKit Agent - Multi-modal agent for real-time geopolitical warning and investigation
How AI Can Enhance The Intelligence Community's Ability To Provide Strategic Warning
The Critical Role of Strategic Warning
Intelligence analysts across the U.S. and UK intelligence communities (ICs) share a common purpose, which is to provide crucial, accurate, and actionable intelligence as quickly as possible so that their political leaders can maintain decision advantage. An essential part of the IC’s mission is to alert leaders to the likelihood and implications of consequential geopolitical events such as imminent acts of military aggression by adversaries, or sudden political transitions, or economic crises. Especially now, U.S. and UK decisionmakers are counting on their intelligence agencies to provide timely and accurate forecasts on the activities of the so-called “axis of disruptors” - China, Russia, Iran, and North Korea – that individually, and increasingly collectively, are attacking neighboring states, undermining democracy, and trying to weaken the Western-led global order.
For policymakers, simply knowing an event of strategic consequence is about to happen is not enough. Policymakers need as much advance warning as their intelligence services can provide them to prepare possible next moves, whether that is implementing a deterrence strategy, rallying allies through diplomacy, or strengthening domestic cyberdefense or defensive economic measures. Adding to the challenge that intelligence services face is collecting and accurately assessing all the relevant data and determining which are the most salient indicators of a brewing crisis. Occasionally intelligence services are able to find a clear-cut nugget of information that reveals an adversary’s intentions, but more often intelligence analysts have to compile a mosaic of data that, when taken together and interpreted correctly, can indicate possible trouble ahead. As the world becomes more and more digitized, the volume of openly-available information is exploding, making it more difficult for analysts to quickly sort through mountains of classified and unclassified information to determine which indicators actually matter, which data to believe, and which to treat with skepticism. Could artificial intelligence (AI) help analysts make this strategic warning system better?
Human Frailty
The IC relies on a human-led approach to intelligence analysis where highly-trained analysts manually weigh several types of intelligence, including human intelligence, signals intelligence (SIGINT), open-source intelligence (OSINT), and imagery, to create an assessment. The benefit of this approach is an assessment that has defensibility and accountability. However, the downsides are also clear. It is time-intensive and susceptible to bias. And there are limits to the amount of data that a human, or a team of humans, can quickly and accurately process, leading to delays or missed insights gleaned from large datasets. In addition, analysts could misinterpret data, misjudge the data’s credibility, and even fail to incorporate relevant data altogether. Furthermore, depending on the training, an analyst may not be skilled in leveraging both quantitative and qualitative data. The result is that warnings of key events stay at the tactical level rather than strategic level.
Advancements in machine learning have provided early evidence to analysts and political scientists that AI could improve on speed and breadth of insights. A growing number of private sector companies are taking advantage of these technological advances to serve client needs. For example, Rhombus Power’s deployment of AI to make reportedly accurate predictions of Russia’s invasion of Ukraine and the U.S. military’s Raven Sentry predictions of Taliban attacks in Afghanistan demonstrate the clear advantages AI could have in making geopolitical forecasts. The two aforementioned models relied on open-source data, which suggests a model that also uses classified data could perform even better.
Leveraging the Power of AI
Admittedly, right now AI tools currently on the market cannot reliably predict a geopolitical event significantly earlier than a human; however, this is likely to change as AI systems mature and become more powerful over the next couple of years. Intelligence leaders should take steps now to prepare their organizations to make use of them. The next generation of AI tools will have the capacity to incorporate far more data - both historical and geographical - and simultaneously reason through several scenarios to make a more comprehensive forecast. We are already seeing AI-based tools used for strategic decisionmaking and forecasts in the financial sector, so their use for geopolitical forecasting is a natural next step. As U.S. and UK intelligence services think about how to integrate AI tools to enhance strategic warning capabilities, some areas they should focus on include:
AI “agents” to monitor the open-source: The IC is often spread thin, which means it cannot always allocate resources, particularly human personnel, to carefully monitor every country and situation. The next generation of AI tools—following in the footsteps of Open AI’s just-released o1 model and Anthropics’s 3.5 Sonnet—will be capable of orchestrating a series of actions in pursuit of broader goals. AI agents that monitor the open-source information space of countries where strategic events are currently not expected would allow the IC to deploy human analysts where they are most needed, such as the axis of disruptors, and reallocate analysts as needed if the agent provides a warning of anomalous activity.
Scenario generation: Often when providing strategic warnings, all-source analysts at agencies like CIA or MI6 are not predicting a singular event but rather assessing a range of possible events and outcomes. When analysts manually generate these scenarios, they are limited by time and to what they already believe to be likely outcomes. Instead, scenario generation could help analysts quickly identify new scenarios they had not yet considered and weigh those scenarios against one another to determine the most likely outcomes.
Data fusion and synthesization: A system that could more effectively and efficiently process various types of intelligence would be highly valuable. One possible use case would be a system that would better fuse crucial SIGINT from NSA with OSINT like public commentary and speeches made by political leaders to detect patterns and attempts to obfuscate. One challenge of strategic warning and geopolitical forecasts is the limited ability to discern or incorporate leadership decisionmaking; adversarial political leaders determine the geopolitical events about which U.S. policymakers want strategic warnings. A better data fusion system would help analysts more comprehensively “get into the heads” of these leaders.
Improvements in human-machine teaming: Tools like crowdsourced human forecasts and automatic human feedback would bring together the rigor and breadth of expertise of human analysts and the speed of these AI tools by positioning humans to be in key positions within the process, including in the development of assessments where human-based explainability is considered imperative.
Why the Time to Act is Now
There are several potential entry points for the IC to integrate AI tools, particularly for intelligence analysis. So why choose investment in a strategic warning system, especially when the technology is still nascent? We argue that there are three key reasons:
Time is of the essence: any tool that can give policymakers a time advantage on crucial topics of national security will always be an upside for the IC’s mission. The time advantage as well as the ability to process more and more data will only improve as the technology develops. Early warnings - even a day’s notice – before a violent event can save U.S. and UK lives or give a bigger window to gather intelligence.
Our adversaries are using it: China’s intelligence apparatus is using every tool in its arsenal, including AI, to challenge our IC, which means a Chinese AI-based strategic warning system could be on the horizon. Integrating AI tools for high impact IC purposes is how we keep our competitive advantage, especially as AI itself becomes part of the threat landscape which necessitates strategic warnings.
Commercial sector development: The development of these tools by technology experts in the commercial sector means that the IC would not need to start from scratch to integrate these tools. These tools would require fine-tuning especially to survive contact with classified or other niche IC data but could be more rapidly integrated into the IC.
In the short to medium-term, analysts will likely use these tools as another source of intelligence rather than rely on the tool as a source of finished intelligence. Even in the long-term, a human analyst will probably need to validate the AI’s system outputs before passing them over to policymakers. However, there is no downside to that because it will allow for consistent fine-tuning, experimentation, and improvement with the goal of generating the quickest, most reliable warnings for decision makers.