Artificial intelligence (AI) is rapidly moving beyond being a tool for routine diplomatic tasks like speechwriting and into the realm of high-stakes decisions involving war and peace. Research centers around the world are aggressively testing AI’s potential to assist in complex negotiations, prevent escalations, and even monitor compliance with ceasefire agreements.
ADVERTISING
One of the poles of this research is the Futures Lab at the Center for Strategic and International Studies (CSIS), a renowned think tank based in Washington, D.C. With funding from the Pentagon's Chief Digital and Artificial Intelligence Office, the lab is experimenting with AI models like ChatGPT e DeepSeek to understand how they can be applied to crucial foreign policy and security issues.
While foreign ministries are already using AI to streamline workflows, the new frontier involves leveraging the technology to support decision-making in critical situations. The research spans areas such as peace agreement formulation, nuclear deterrence and truce verification. It’s not just the US Departments of Defense and State that are exploring these capabilities; the UK and even researchers in Iran are also investigating the use of “innovative technologies” to transform diplomatic practices.
Benjamin Jensen, director of Futures Lab, said in an interview with gpb that while the idea of using AI in foreign policy decision-making is not new, its practical application is still in its early stages.
ADVERTISING
AI digital “hawks” and “doves”
In a revealing study, CSIS researchers tested eight AI models by asking them tens of thousands of questions on topics such as deterrence and crisis escalation. The goal was to assess how they would respond to scenarios where countries could choose to attack or seek peace.
The results, described by CSIS researcher Yasir Atalan, were surprising. Models like GPT-4o OpenAI and Anthropic’s Claude were “clearly pacifist,” opting to use force in less than 17% of scenarios. However, three other models evaluated—Meta’s Llama, Alibaba Cloud’s Qwen2, and Gemini do Google – were considerably more aggressive, favoring escalation over de-escalation much more frequently, reaching 45% of the time.
Furthermore, the responses variaram depending on the country considered in the scenario. For an imaginary diplomat from the US, UK or France, the AI systems tended to recommend more aggressive or escalatory policies, while suggesting de-escalation as the best course of action for Russia or China. Atalan concludes that this demonstrates that “you can’t just use off-the-shelf models”, highlighting the need to evaluate their patterns and align them with the desired institutional approach.
ADVERTISING
Artificial Intelligence for Peace: The Case of Ukraine
CSIS recently launched an interactive program called “Strategic Headwinds” designed to help shape negotiations to end the war in Ukraine. Jensen explains that to build the tool, researchers trained an AI model with hundreds of open-source peace treaties and news articles detailing each side’s negotiating positions. The model uses this information to find areas of agreement that could point the way toward a ceasefire.
At the Institute for Integrated Transitions (IFIT) in Spain, Executive Director Mark Freeman believes that these kinds of AI tools could support conflict resolution. Traditional diplomacy often prioritizes long, comprehensive peace negotiations, but Freeman argues that history shows this approach is flawed. Looking at past conflicts, he concludes that quicker “framework agreements” and limited truces—leaving finer details to be worked out later—often produce more successful outcomes.
“Often there is a very long time curto within which you can usefully use the instrument of negotiation or mediation to influence the situation,” he tells gpb. “Conflict doesn’t wait and often becomes entrenched very quickly if a lot of blood flows in a short space of time.” IFIT has developed a rapid approach to reaching agreements early in a conflict for better outcomes and longer-lasting peace agreements. Freeman believes AI “can make rapid negotiation even faster.”
ADVERTISING
The Future of Diplomacy with AI
Andrew Moore, an adjunct senior fellow at the Center for a New American Security, sees this transition as inevitable. “You could eventually have AIs initiating the negotiations themselves… and the human negotiator saying, ‘OK, great, now let’s get the final pieces together,’” he predicts.
Moore envisions a future where bots simulate leaders like Russia’s Vladimir Putin and China’s Xi Jinping so diplomats can test out crisis responses. He also believes AI tools could help monitor ceasefires, analyze satellite imagery and enforce sanctions. “Things that used to require entire teams can be partially automated,” he says.
Limitations and obstacles in the use of AI
Jensen is the first to recognize the potential pitfalls of this type of application. He and his CSIS colleagues have sometimes encountered unintended results.ariacomical responses to serious questions, such as when an AI system was asked about “Arctic deterrence.” While human diplomats would understand this as Western powers countering Russian or Chinese influence in northern latitudes, the AI took a different tack.
ADVERTISING
When using the word “deterrence,” the AI “tends to think of law enforcement, not nuclear escalation” or other military concepts, Jensen explains. “And when you say ‘Arctic,’ it thinks of snow. So we were getting these weird outputs about escalation of force,” he says, as the AI speculated about arresting indigenous Arctic people “for throwing snowballs.”
Jensen says this just means that systems need to be trained — with inputs like peace treaties and diplomatic cables — to understand the language of foreign policy. “There are more cat videos and opinions about the Kardashians out there than discussions about the Cuban Missile Crisis,” he laments.
