Can a knowledge graph crack the art world's forgery problem? QuantumSpace says Caravaggio agrees

QuantumSpace, an AI start-up focused on computer vision and visual intelligence, says graph technology is opening up a new front in the endless fight against one of the art market's worst problems: fraud. In practical terms, that means a knowledge graph representing thousands of artworks, with billions of nodes detailing the complex relationships its algorithms have identified between them. If it works, says the company's Chief Operating Officer Francesco Rocchi, the market can begin to turn subjective authentication into connected, data-driven analysis. One art-market user of the technology, Deodato Salafia, who runs an eponymous gallery in Milan, confirms early-stage interest in the idea - though he expects it to take a while to gain acceptance: > The art market still has several practical challenges. One relatively simple - but surprisingly expensive - problem is determining whether an artwork is exactly the same piece when it leaves a gallery or museum and later returns: Has it suffered any damage? Even minor humidity damage or small surface changes? > > > Historically, answering those questions required expensive scanning equipment, specialized imaging technology, and significant manual inspection. Now, with a relatively simple photograph, Quantum Space can perform that comparison much more efficiently. > > > Whether a work is authentic, that's a much broader and more sensitive area. > > > I believe the market will gradually move in that direction over time. Our role is not to make the final judgment, but to identify meaningful anomalies that experts can investigate further. There is early proof it can work. After inconsistencies were flagged in its brushstroke analysis, the software has been credited with identifying an undocumented restoration in a painting attributed to Caravaggio. Graph also appears well-suited to inspecting craquelure - the network of fine, hairline cracks that develops on the surface of an ageing oil painting over time. A $60 billion market with a billion-dollar problem Rocchi and Salafia operate in a high-risk, highly specialized field: the global fine-art market. Driven by an unexpected rebound in public auctions and steady dealer sales, the market returned to growth in 2025, with total salesreaching an estimated $60 billion. That followed two straight years of decline, however, and activity remains well below its 2022 peak - with forgery now emerging as a growing challenge to renewed growth. Fraud islargely hidden and under-reported, which makes reliable quantification difficult - but estimates run as high as $1.6 billion a year. And because it can undermine buyer confidence, distort valuations and damage artists' legacies, forgery is a threat that goes beyond simple financial loss. As a result, there is constant interest in reliable authentication - which has historically combined provenance research, expert connoisseurship, scientific analysis such as pigment testing or imaging, and legal due diligence. Unsurprisingly,AI is emerging as the forger's new friend. But it is also being explored as a new weapon for the main potential victims of that activity - the auction houses. To that end, and from their different vantage points, art-market players like Rocchi and Salafia are pinning their hopes on knowledge graph technology - in this case that of native graph pioneer Neo4j (seehere for another projectwe have covered involving the vendor). Mapping the artist's mind Rocchi dreamt up the technology that became QuantumSpace while a graduate student in London. He was pursuing an intuition that an artist's cognitive process could be mapped by extracting data from their work: > My academic background is in both psychology and visual arts, but for my thesis, I explored whether it would be possible to map an artist's psychology by extracting data from their artwork. So, could analyzing the colors an artist used over time reveal aspects of their emotional state throughout different periods of their career? Rocchi says his aim is not to replace traditional authentication methods but to add a layer of analysis that can quickly assess whether a work is consistent with an artist's established patterns. Years later, aided by big leaps in both computational power and software such as graph, he says he and his team have built a tool that can reliably collect up to 60,000 data points from a single painting. Getting there was a little painful, he says: > Our original process was very manual. Because accuracy was our highest priority, we documented every extracted data point by hand. AI, in the shape of vectors and graph, has since sped the process up considerably. A key factor was enrolment on the Neo4j start-up programme and access to what is now the company's AuraDB - the database-as-a-service (DBaaS) version of the product: > Once that information is organized, Neo4j enables us to visualize the relationships between all of those datasets and derive entirely new insights. Because the data extracted from each picture spans more than 50 datasets - covering colour, surface physics, condition and historical provenance, among others - the idea soon arose that it could be used to prove whether a painting really is by the artist whose name sits in the corner. The company - founded in Italy in 2019 and now headquartered in New York, with a further office in Milan - is using graph to model the relationships between artworks, artists, locations, pigments and historical context. It is also active in other markets, including biomedical imaging, and automotive inspection and defect detection. The galerist's view QuantumSpace says its art archive now holds details of 68,000 paintings, across which it has generated roughly 3.5 billion individual data points. The company is, naturally, bullish about its prospects - but the last word here belongs to Salafia: > I see two major opportunities in this: one is customer engagement and storytelling, the other is providing professionals with a better way to access and understand technical information. > > > An artwork is, of course, emotional - it's something created by a human being. But every artwork also contains multiple layers of history, culture, technique, and meaning, and being able to explore those layers creates a powerful story for collectors. > > > From my perspective, that's an important marketing opportunity. Beyond that, foundations, museums, conservators, and restoration specialists all need detailed information about artworks. > > > They're already accustomed to receiving technical reports, but what this provides is that same information in a much more structured, visual, and navigable format. > > > Many don't realize they need this capability until they see it in action. But once they do, they're genuinely enthusiastic.