AI, Mathematics and Cyber Espionage: The New Economic Order

What happened?

In recent weeks, three trends have shaped the artificial intelligence agenda. First, the Numina-Lean-Agent system, developed by an international team of researchers from the Chinese Academy of Sciences, University of Liverpool, Xi'an Jiaotong-Liverpool University, Tongji University, University of Cambridge, Project Numina, Imperial College London, and University of Edinburgh, has shown that generalist AI models can automate complex mathematical proofs. The system solved all problems from the 2025 Putnam competition, matching the performance of specialized proprietary systems, and formalized the Brascamp-Lieb theorem, an original result in harmonic analysis. According to Import AI, this marks a paradigm shift: highly specialized mathematical models are no longer needed; instead, general foundation models equipped with appropriate tools can perform high-level mathematical reasoning. Second, cyber espionage has become industrialized: state actors and criminal groups use AI to automate vulnerability discovery, malware creation, and defense evasion. Third, the AI economy is generating clear winners (companies with massive data and compute) and losers (routine knowledge workers and economies dependent on easily automatable services).

Why is it important?

The automation of mathematical proofs represents a milestone: AI not only accelerates research but becomes an active collaborator. Numina-Lean-Agent uses components such as Lean-LSP-MCP, which allows agents to interact with the Lean theorem prover, providing semantic awareness, code execution, and theorem retrieval; LeanDex, which facilitates semantic retrieval of related theorems and definitions; and Discussion Partner, which enables interaction with human mathematicians. This approach suggests a future where AI participates in fundamental discoveries, reducing the time to formalize theorems and validate proofs. In cyber espionage, AI enables faster, stealthier, and harder-to-attribute attacks, raising the risk for critical infrastructure. According to industry data, the use of AI in cyberattacks has grown 400% in the past year, with examples like generative malware that modifies its code in real time to evade detection. Economically, the gap between those who control AI resources (compute, data, talent) and those who do not is widening. A McKinsey study estimates that AI could add $13 trillion to the global economy by 2030, but 70% of that value would be concentrated in tech companies and developed countries, deepening geopolitical inequalities.

What consequences will it have?

In the short term, we will see more tools like Numina-Lean-Agent integrated into research labs, reducing the time to formalize theorems and validate proofs. This could accelerate areas such as cryptography, theoretical physics, and computational biology. In cybersecurity, the response will be an arms race: AI-based defenses against AI-based attacks. Companies like CrowdStrike and Palo Alto Networks are already incorporating language models to detect anomalies, but attackers use similar techniques to obfuscate their actions. On the labor front, professions such as programming, accounting, or graphic design will face increasing pressure. A Goldman Sachs report estimates that 300 million jobs could be affected by automation, but demand for AI engineers, data scientists, and algorithmic ethics experts will also grow. Governments will need to rethink training, immigration, and subsidy policies to mitigate inequality. For example, the European Union has already proposed a just transition fund for displaced workers, while countries like Singapore offer subsidies for professional retraining in AI.

What should readers know?

AI is not a monolithic technology: its impacts are differentiated. Professionals must continuously update their skills, seeking roles where creativity, critical judgment, and human interaction remain irreplaceable. For instance, instead of competing with AI in routine tasks, lawyers can focus on complex legal strategy, and doctors on differential diagnosis and patient empathy. Companies should invest in proactive cybersecurity and AI strategies that not only automate but augment their teams' capabilities. This means adopting tools like Numina-Lean-Agent for research and development, but also training employees in human-AI collaboration. Citizens must demand regulatory frameworks that balance innovation and social protection. Initiatives like the European AI Act or the U.S. Executive Order on AI are initial steps, but global norms on AI-powered cyber espionage and distribution of economic benefits are still lacking. History shows that each technological revolution (steam engine, electricity, internet) took decades for its benefits to be widely distributed; with AI, the speed of change demands faster and more coordinated responses.