CIA builds digital twins of world leaders to predict moves by Putin and Xi Jinping

The CIA has built an AI tool that assembles "digital twins" of foreign leaders and lets analysts use them as a simulator. The system took about two years to develop. Its main purpose is not automated diplomacy. Instead, it aims to predict more accurately how a specific president or prime minister will behave in a crisis, at the negotiating table, or under pressure.

What the CIA built

The tool is a chatbot that the US intelligence community uses as a model of a specific politician. Each "twin" is supposed to answer questions, argue, explain decisions, and respond to prompts the way analysts believe its real-world counterpart would. Staff from foreign intelligence and counterintelligence worked on the project, and its goal comes down to one thing: turning scattered knowledge about a leader into a tool analysts can engage with in conversation, not just through briefs and dossiers.

Judging by the description, this is not an "avatar" or a deepfake but a behavioral model. It likely draws on public speeches, biography, negotiating style, past decisions, reactions to crises, inner circle, and political context. The idea is that an analyst doesn't just read through an archive but can ask the system something like "how will this leader respond to a new sanctions package" and get an answer built from the accumulated body of data rather than random text.

How it will be used

The core point of such models is to run scenarios before they happen in reality. For intelligence agencies, this is a convenient way to test hypotheses, compare pressure options, and judge which signals are more likely to produce a concession, an escalation, or a prolonged standoff. The tool is especially useful where the cost of a mistake is high and real opportunities to "practice" barely exist.

• Modeling reactions to sanctions, ultimatums, and diplomatic proposals
• Testing a leader's behavior in crisis negotiations
• Assessing likely rhetoric for domestic and foreign audiences
• Comparing multiple escalation and de-escalation scenarios
• Preparing analytical memos for decision-makers in Washington

Still, a bot like this cannot be an oracle. Any model depends on the quality of its data and on the assumptions of the people who tune it. It works well on recurring patterns but does worse at catching impulsive decisions, hidden agreements, and factors absent from open sources. If analysts start trusting the simulation too much, they risk mistaking a convenient picture for real politics.

Why Putin and Xi matter

The particular interest in the profiles of Vladimir Putin and Xi Jinping makes sense. They lead countries whose decisions directly affect war and security, supply chains, energy markets, technology restrictions, and the architecture of international alliances. For the US, any tool that promises even a slightly better grasp of their reasoning automatically becomes strategically important, even if the accuracy of such forecasts is far from ideal.

There is another reason. The less transparent decision-making at the top of a state, the more valuable a model that tries to reconstruct a leader's thinking from indirect signals. With Russia and China, outside observers often see only official statements and outcomes, not the deliberation behind them. So the digital twin here serves not as a "truth machine" but as a working testing ground for hypotheses, assumptions, and intelligence assessments.

What it means

AI is visibly shifting from a tool for generating text and code into an applied layer for government analysis and geopolitical modeling. The digital twins story shows that the next round of AI competition is not only about the most powerful models but also about the ability to build narrow, domain-specific simulators for concrete decisions — from business to intelligence.