Tokyo's Duplicate Image Crisis: The Key Decisions That Will Shape How the City Fixes Its Digital Records

Tokyo's ward offices are sitting on a problem they can no longer quietly manage. Across at least a dozen of the capital's 23 special wards, municipal digital archives — covering everything from building permit photographs to resident registration documents — contain duplicate, mislinked, or incorrectly assigned images that have accumulated over more than a decade of piecemeal digitisation. The question now is not whether to fix it, but how fast, at what cost, and under whose authority.

The issue has sharpened in 2026 because of competing pressures: a record inbound tourism surge straining ward-level administrative systems, an ongoing yen weakness that has made imported software licensing more expensive, and a national push under the Digital Agency's Digi-Den (デジタル田園) programme to have all core local government records migrated to standardised cloud infrastructure by March 2027. That deadline has forced ward offices to audit what they actually have before they move it — and audits have surfaced the duplicate image problem at scale.

What the Problem Looks Like on the Ground

In Shinjuku Ward, whose civil affairs division manages records for roughly 346,000 registered residents, internal review processes identified image duplication across property inquiry files processed between 2013 and 2021 — a period when the ward ran parallel paper-scanning workflows and a separate vendor-supplied digital intake system that did not consistently cross-check file identifiers. Koto Ward, which has seen heavy redevelopment pressure around the Ariake and Tatsumi districts, faces a related problem in its building confirmation photograph archive, where images from multiple inspection visits were sometimes stored under a single application number.

Neither ward has made a public statement about the scale or cost of remediation. But the Digital Agency's national framework, updated in April 2026, now requires local governments to submit a data-quality declaration before any cloud migration can be approved. That declaration must certify that duplicate or orphaned records have been identified and either corrected or flagged. For wards that have not yet started that process, the March 2027 deadline is effectively a forcing function.

Vendors are circling. At least three major Japanese systems integrators — including Fujitsu and NTT DATA, both of which hold existing contracts with Tokyo Metropolitan Government — have been marketing AI-assisted deduplication tools since early this year. A pilot run in Sumida Ward, using image-hash comparison software to scan roughly 180,000 archived permit photographs, reportedly took six weeks and flagged about 4,200 files for human review. Whether that cost and timeline is acceptable across all 23 wards at once is a core decision still unmade.

The Decisions That Matter Most

Three choices will define what happens next. First, centralisation versus ward autonomy: Tokyo Metropolitan Government, whose headquarters sit in Nishi-Shinjuku, could take over the deduplication exercise as a single metro-wide project, pooling contracts and standardising methodology. Alternatively, each ward could procure independently, as they have historically done for IT systems — a path that risks 23 different definitions of what counts as a duplicate. The metropolitan government has not yet indicated which model it prefers.

Second, AI sign-off authority: even where automated tools flag duplicates reliably, someone must authorise deletion or merging of records that may have legal standing under Japan's Gyōsei Bunsho Kanri Hō (行政文書管理法), the administrative documents management law. Current rules require a human official to authorise destruction of any government record. Clarifying whether AI-generated deduplication constitutes a preliminary sort or a final determination is a legal question the Internal Affairs Ministry has not yet resolved publicly.

Third, cost allocation: with the yen trading near multi-decade lows against the dollar, dollar-denominated cloud storage and American-built AI licensing costs are materially higher than they were when initial digitisation budgets were set. Wards with smaller rate-base revenues — including Arakawa and Kita — are watching to see whether the metropolitan government or the national Digital Agency will provide remediation subsidies.

The practical calendar is unforgiving. Ward offices that want to meet the March 2027 cloud migration deadline need to submit their data-quality declarations by November 2026 at the latest to leave time for correction cycles. That means procurement decisions for deduplication tooling need to happen before September — giving officials roughly ten weeks to answer questions that have sat unanswered for years.