A former head of state with a background in physics has called for a binding international agreement to govern artificial intelligence — arguing that no single nation can manage the risks of rapidly advancing AI technology alone, and that the window for action is narrowing.
A Scientist-Turned-Statesman Speaks on AI Risk
Petre Roman, former Prime Minister of Romania and a physicist by training, joined Sanjay Puri on the RegulatingAI Podcast — recorded live at the Club de Madrid Annual Policy Summit — to share his perspective on the governance challenges posed by artificial intelligence. Drawing on both scientific understanding and decades of political leadership, Roman made a direct case for international regulatory cooperation.
Roman warned that frontier AI models are already demonstrating the capacity to identify vulnerabilities in critical infrastructure, and that voluntary corporate restrictions on powerful models offer only temporary protection since similar systems will inevitably emerge elsewhere. “Nobody can really stop AI development,” he acknowledged. “What we need is a global treaty on that.” He framed the current moment as analogous to the nuclear age — a time when no individual country could responsibly manage transformative technology without multilateral coordination.
The Transparency Imperative
One of Roman’s strongest arguments centered on transparency as an essential foundation for AI governance. He called on technology companies to clearly label AI-generated content so users can distinguish it from human-created material, and to be more open about how their models are built and deployed. “Transparency is perhaps the essential part of AI regulation,” he stated during the conversation.
The Risk of Plausibility Over Truth
Roman raised a particularly pointed concern about how many current AI systems are designed to maximize user satisfaction rather than factual accuracy. He warned that these systems “stimulate plausibility much more than the truth” — a tendency that, at scale, could fundamentally distort public understanding of reality. He also cautioned against over-reliance on AI chatbots, noting that while they can address loneliness and offer practical assistance, they cannot replicate human feelings, intuition, or the capacity to distinguish between right and wrong.
Water, Energy, and Environmental Concerns
Roman brought an unexpected dimension to the AI policy conversation by flagging the environmental implications of rapidly expanding data center infrastructure. While renewable energy adoption can address electricity demand, he argued that water consumption for cooling poses a more serious long-term challenge — one that affects local communities and ecosystems in ways that are often overlooked in policy discussions. He called on AI companies to invest in more energy-efficient training methods and sustainable infrastructure, and argued that local communities should have a meaningful voice in decisions about major AI infrastructure projects.
Protecting Democracy in the Age of AI Disinformation
Drawing on Romania’s experience with AI-driven disinformation campaigns during its 2024 elections, Roman expressed serious concern about artificial intelligence’s growing capacity to undermine democratic institutions by distorting public debate and influencing voters. He stressed that democracy depends on citizens having access to reliable information and urged governments to act decisively before the risks become harder to contain. His closing message struck a balance between urgency and optimism: “We have to regulate artificial intelligence without stopping its very beneficial development.”
