Can AI End the War and Strengthen Democracies Against Autocracies?

Summary

The discussion examined whether AI can help end the war in Ukraine and strengthen democracies against autocracies. The moderator opened by emphasizing the human cost of the war and cited recent reports of AI misuse, including cyberattacks and attempts to support harmful activities, as evidence that the risks are both immediate and long-term.

AI on the Battlefield

Artem Martynenko identified three main military applications of AI:

• Processing large volumes of sensor and battlefield data
• Helping drones operate in GPS-denied or electronically contested environments
• Accelerating the development of software and equipment

He described Ukraine’s Delta system, which combines battlefield information and uses AI to help analyze drone video and identify enemy activity. Ukraine’s video-processing tool uses a locally deployed model based on open-source technology rather than a public commercial model.

Martynenko argued that AI is a tool, not a decisive advantage on its own. Its value depends on people, processes, and technology working together. Since Russia is developing similar capabilities, AI may not give either side a lasting edge. In his view, Ukraine’s more horizontally connected style of communication has helped it make use of systems such as Delta.

Louis Mosley also cautioned that AI is not a magic solution to war. It can change the cost and efficiency of military operations and may be especially valuable to Ukraine by reducing reliance on scarce human resources. However, tactical gains do not guarantee strategic success: military effectiveness still requires a clear theory of victory.

The speakers emphasized that people remain central, even as increasingly autonomous systems take on parts of the military observe–orient–decide–act cycle. They said doctrines for delegating authority to human commanders will need to be considered as AI assumes more functions.

Cybersecurity and AI Misuse

Martynenko described cyberattacks as a daily reality for Ukraine, affecting military systems, the economy, and public services. Threats range from attempts to compromise individual computers to social engineering targeting soldiers and their families, as well as the use of advanced models to plan or support complex cyber operations. He said Ukraine’s defense technology work does not use Chinese models.

AI Governance and Democratic Cooperation

Annette Kroeber-Riel praised Ukraine’s rapid adoption of digital and AI tools beyond the battlefield, including services supporting veterans and education during wartime. She supported a proposal by Google DeepMind chief Demis Hassabis for a rapid oversight mechanism for advanced AI. She argued that safety testing, audits, and cooperation between governments and industry need to keep pace with development, with pauses possible if systems cannot be adequately understood or assessed.

Kroeber-Riel said democracies should cooperate and share information, while acknowledging that a genuinely global governance framework involving autocracies may not be realistic at present.

Sanna Marin described the tension between AI’s potential benefits and the risk of severe harm if autonomous systems make life-and-death decisions without effective safeguards. She called for international rules involving both governments and companies, and warned that technologies developed in the context of war may shape their broader use.

The discussion also highlighted the challenges of regulating AI amid strategic competition. The moderator asked whether democratic safeguards could work if China did not follow them. Mosley argued that the decisive competition is concentrated in Washington and Beijing, leaving Europe with limited influence over the development of frontier systems. He compared the strategic stakes to past nuclear competition, questioned whether AI companies’ warnings about existential risks also serve commercial interests, and urged attention to the wider technology supply chain—including chips and the specialized equipment needed to produce them.

The moderator concluded that even an imperfect agreement on AI misuse would be preferable to having no controls, while noting that verification may be difficult.

The session ended without audience questions because of time constraints.

Presenters and Contributors

Niall Ferguson (moderator), Artem Martynenko, Louis Mosley, Annette Kroeber-Riel, and Sanna Marin.