Madrid Archivists Split on AI Tools That Restore Damaged Public Records

A quiet but intensifying dispute has broken out among Madrid's archivists, urban planners and digital preservationists over the growing use of artificial intelligence tools that automatically detect and replace damaged or duplicate images embedded in public records, cadastral databases and cultural heritage files. The core question is blunt: when a software system substitutes a corrupted photograph with a reconstructed or sourced replacement, does the official record remain trustworthy?

The debate has sharpened this spring because the Comunidad de Madrid's Archivo Regional de la Comunidad, based on Calle Ramírez de Prado in the Arganzuela district, is midway through a multi-year digitisation drive that began in 2023. Staff there are cataloguing tens of thousands of analogue documents, many of which contain photographs that have faded, torn or duplicated across different administrative files. Automated tools offered by at least three European software vendors are being evaluated to accelerate that process. Critics inside the profession say the evaluation is happening faster than the ethical frameworks governing it.

What the Experts Are Actually Saying

Archivists and records professionals contacted for this article described a profession split roughly along generational lines. Older practitioners, many trained under Spain's Cuerpo Facultativo de Archiveros, Bibliotecarios y Arqueólogos framework, argue that any alteration to an original document image, even replacing a visibly corrupted file, must be logged with full provenance metadata, traceable to a named human decision. Younger digital specialists counter that the tools in question do not delete originals; they layer replacements transparently, the way a conservator applies reversible materials to a damaged canvas.

The Colegio Oficial de Bibliotecarios y Documentalistas de la Comunidad de Madrid has been asked by its membership to clarify its position on AI-assisted image replacement in public records, according to information circulating at a professional seminar held at the Biblioteca Regional Joaquín Leguina in Retiro in late May. No formal policy statement had been published as of this article's filing date.

The argument is not purely theoretical. Madrid's urban planning directorate, the Área de Desarrollo Urbano, maintains a vast image database tied to its PGOU, the Plan General de Ordenación Urbana, which contains thousands of site photographs attached to building permits and zoning decisions dating to the 1990s. File corruption and duplication are documented problems in datasets of that age and scale. Officials within the directorate have acknowledged internally that some permit records contain images that no longer correspond to the correct properties, though the scale of the problem has not been officially disclosed.

The Practical Stakes in a City Under Scrutiny

Madrid is under particular pressure to get this right. The city's rental market, already strained by the ongoing controversy over the regional government's resistance to rental caps backed by opposition parties in the Asamblea de Madrid, depends on accurate cadastral and planning records. A duplicated or incorrectly replaced property image in a public database can affect valuations, licensing decisions and even eviction proceedings. Housing lawyers operating in Lavapiés and Carabanchel, two neighbourhoods with high concentrations of rental disputes, have flagged to local legal associations that image discrepancies in administrative files occasionally surface during court challenges.

The European Union's AI Act, which entered into force in August 2024 and whose high-risk provisions are being phased in through 2026 and 2027, adds regulatory weight to the conversation. Systems that automatically modify documents used in administrative or judicial processes may fall under the Act's transparency and human-oversight requirements, depending on how they are classified. Madrid's archivists are waiting for clearer national guidance from Spain's Agencia Española de Supervisión de la Inteligencia Artificial, which is still developing sector-specific criteria.

For institutions evaluating these tools right now, specialists recommend three immediate steps: ensure all automated replacements are stored as a separate version layer rather than overwriting the original file, require a named staff member to authorise each batch replacement rather than running processes overnight without review, and log every substitution in a change registry that is itself part of the public record. The Archivo Regional's current evaluation phase is expected to produce a recommendation by the end of September 2026, leaving Madrid's cultural heritage community a narrow window to shape the standards before decisions are locked in.