Arlequin AI raises €28M Series A for topological neural networks

Why it matters

Arlequin AI is testing whether explainable, relationship-aware models can win high-stakes enterprise and defense work without matching frontier labs' compute spending.

Hugo Micheron and Antoine Jardin have raised €28 million for Arlequin AI in a Series A reported Thursday by Tech.eu. The Paris company is developing topological neural networks for analyzing relationships across large, fragmented datasets.

Redalpine and OTB Ventures co-led the round. Bpifrance's Defence Innovation Fund participated, while existing backers Vsquared Ventures and 10x Founders increased their stakes. Arlequin AI did not disclose a valuation or the investors' individual check sizes.

The financing follows a €4.4 million seed round announced in June 2025 and gives the two-year-old Paris lab considerably more room to develop what it calls topological neural networks. Arlequin AI says those models can analyze documents, transactions, video and operational records as a connected system, preserving a path from each result back to the underlying evidence.

That traceability is central to Arlequin AI's pitch. Micheron and Jardin are selling to organizations where an authoritative-sounding fabrication can carry legal, financial or national-security consequences. Tech.eu reported that governments and large organizations across Western and Eastern Europe already use Arlequin AI, although customer counts, contract values and current revenue were not disclosed.

A political scientist's architecture bet

Micheron's path into AI began with field research rather than software engineering. He studied international relations at Sciences Po Aix, learned Arabic during a year in Syria in 2008 and 2009, and completed a Middle East-focused master's degree at King's College London. His doctoral work, completed at the Ecole normale superieure in 2019, grew out of research into Europeans leaving for Syria and included interviews with imprisoned jihadists.

Micheron later taught at Princeton University before returning to Sciences Po in 2023. His second book on European jihadism won France's Prix Femina for nonfiction that year. In a Sciences Po interview, he described jihadism as a way to examine the dynamics running through European societies, including the influence of social networks, artificial intelligence and foreign interference.

Jardin brought the quantitative half of that research partnership. A political scientist and former CNRS research engineer, Jardin has studied political attitudes, urban segregation, insecurity, religion and violence. He has also taught quantitative social-science methods at Sciences Po and the University of Versailles Saint-Quentin-en-Yvelines.

The pair founded Arlequin AI in 2024. Micheron's documented academic focus includes political violence and information environments, while Jardin's work covers human behavior and quantitative methods. Arlequin AI's product thesis is that institutions will pay for software that maps the structure connecting records and shows its work. Conventional search can retrieve individual records, while generative models can summarize them.

Arlequin AI calls its platform HuDEx, or Human Data Explorer. The platform is positioned for security and defense analysis, criminal investigations, fraud and money-laundering detection, cybersecurity, information integrity and AI safety. Earlier Arlequin AI material described HuDEx as an unsupervised, non-generative system built without prompts or pretrained large language models.

Micheron told Tech.eu that a new class of systems must understand "highly complex dynamics hidden within millions of data points." Jardin's technical thesis is similarly direct: further progress will require different architectures instead of repeatedly increasing model size, training data and compute.

The benchmarks now matter

Topological deep learning is a real research field, with open-source implementations and an ICML workshop challenge devoted to architectures that operate on richer relationships than ordinary pairwise graphs. Arlequin AI's commercial and technical advantage remains a company assertion, however.

Arlequin AI says its models can represent higher-order interactions among several elements at once and require less compute than large-scale generative systems. Arlequin AI has not attached public performance benchmarks, customer accuracy data or a peer-reviewed architecture paper to those claims. The Series A is funding the work needed to turn an academically credible premise into a measurable product advantage.

The competitive bar is already established. Quantexa sells graph analytics and entity-resolution software for financial crime, fraud, government and supply-chain investigations. Paris-based Linkurious built graph visualization tools for investigative work before Nuix completed its acquisition in 2026. Arlequin AI is attempting to push further down the stack by owning the models that identify and rank the relationships, rather than serving primarily as an interface for exploring a graph.

Arlequin AI's focus on explainability may prove as important as raw model performance. Banks, intelligence services and investigators need to defend how a conclusion was reached. A system that can point back to a transaction, passage or operational event has a clearer route into those workflows than one offering a fluent answer assembled through opaque reasoning.

The participation of Bpifrance's defense-focused fund also places Arlequin AI inside France's effort to finance dual-use technologies with military and civilian applications. That backing fits Arlequin AI's security use cases and its emphasis on European control over sensitive data processing.

Silicon Valley comes after London and Berlin

Arlequin AI will use the Series A to recruit researchers and engineers, train proprietary models and expand commercial deployment. Arlequin AI has opened offices in London and Berlin and plans to establish a Silicon Valley AI lab in the coming months, extending a European sovereignty pitch into the center of the US model economy.

The Bay Area move would place Arlequin AI beside labs pursuing the scale-heavy strategy the founders reject. Arlequin AI will still need substantial technical resources to prove that alternative mathematics can deliver better economics. Smaller compute requirements remain valuable only if the resulting models perform reliably on the investigations and operational decisions customers care about.

The round buys Arlequin AI time to run that test. Micheron and Jardin have set out an architectural thesis: the relationships inside data deserve as much attention as the data points themselves. Investors have now put €28 million behind the proposition that institutions making high-stakes decisions will agree.