The Agentic AI Archiving Gap — No Current Digital Solution Exists, and Why Hybrid Non-Digital Formats Are the Only Defensible Answer for 25+ Year Preservation
Can digital-only systems preserve records authentically for 25 years or more in the age of AI?
No validated standard currently exists for authenticating records created or modified by agentic AI systems over 25-year timeframes. The world's leading preservation research body, InterPARES Trust AI (2021–2026), confirms the problem remains unsolved. A hybrid architecture anchored by archival microfilm — rated LE 500 under ISO 18901 — is the only currently defensible permanence strategy.
The Core Problem
The archival science community is confronting an irresolvable paradox: AI is now the only tool capable of managing the volume of digital records being created — yet AI simultaneously generates, modifies, and routes records in ways that no current digital standard can authenticate over long timeframes. The world's leading research consortium on this question, InterPARES Trust AI (University of British Columbia, 2021–2026), titles its foundational paper "Trusted Data Forever: Is AI the Answer?" — and after five years of multinational, multidisciplinary investigation, the answer remains: not yet, and possibly not alone. The issue exists for both digital and digitised records and is rendered more serious by the sheer number and volume of records that have accumulated over time and are being created today in a large variety of systems. No purely digital solution for the agentic AI archiving problem currently exists in published, validated form. arxiv
Finding 1: There Is No Current Digital Standard for Authenticating AI-Generated and Agent-Created Records
The authoritative statement comes directly from the archival science community's own literature. A 2025 paper in the Proceedings of the Association for Information Science and Technology (ASIS&T) — one of the field's premier peer-reviewed venues — confirms that researchers have begun to evaluate the evidential properties of AI-generated archives and strategies for evaluating the authenticity and reliability of AI-generated archives, but this work indicates a need to investigate questions pertaining to the evidential foundations and societal impacts of AI-generated and mediated archives. These issues are present at multiple stages of the archival lifecycle — records creation, access, and evaluation and use. Journal of the Association for Information Science and Technology
The 2025 Nordic Archivaria journal paper from InterPARES Trust AI (the project's own researchers) is equally explicit: the archivist's traditional knowledge remains a necessary foundation for the preservation and authentication of digital records, especially the concept of record and the methodologies of paleography and diplomatics. Decisions made and actions taken by AI-enabled systems or tools must be documented, and the temporal dimension of AI agent activity must be captured. "Must be documented" is a prescription for future practice — not a description of a solved problem. Scandinavian University Press
What this means: As of 2026, no ratified, published standard exists for establishing the chain of custody, authenticity, or long-term trustworthiness of records that were created, appraised, modified, or routed by agentic AI systems. The research community is still describing what the documentation requirements should be. They have not yet been standardised, tested, or validated at scale.
Finding 2: The Foundational Archival Standard for Long-Term Preservation Was Revised in 2024/2025 Precisely Because the Digital Framework Was Inadequate — Its Third Major Revision in 22 Years
ISO 14721 — the OAIS Reference Model — is the bedrock standard every trusted digital repository claims compliance with. It defines "Long Term" as long enough to be concerned with the impacts of changing technologies, including support for new media and data formats, and changes within the Designated Community. Long Term may extend indefinitely. ISO
The standard required its third major revision (2003 → 2012 → 2025) specifically to address what the previous versions could not handle. It is not widely believed that there will be a single solution to digital preservation, or that solutions can be achieved solely through technological means. This statement — from a foundational US Congressional-commissioned report on digital preservation principles — has never been contradicted in the archival literature. It was true in 1995 when first written. It remains true in 2026, with agentic AI having made the problem categorically harder. arxiv
What this means: A standard that requires revision every decade is, by definition, not a 25-year solution. Every organisation that built its digital preservation programme to OAIS v1 (2003) needed to re-evaluate by 2012. Every organisation that rebuilt to OAIS v2 (2012) needed to re-evaluate in 2024/2025. The "25-year archive" requires, at minimum, two complete standard-level re-evaluations before it matures. That is not preservation — that is perpetual migration with no guaranteed endpoint.
Finding 3: Hybrid Archives — Physical and Digital Together — Are Already the Practitioner Consensus
The shift from debating "digital vs. analogue" to accepting hybrid as the necessary architecture is visible in the academic literature. A 2025 paper in AI & Society (Springer, DOI: 10.1007/s00146-025-02374-y) documenting cutting-edge AI archival work at the British Library explicitly describes the research focus as managing correspondence collections in hybrid form — paper and digital — noting that the Harold Pinter Archive project uses AI tools to manage the relationship between 20,000 paper letters and 3,500 emails. The collection-agnostic code enables GDPR-compliant metadata extraction and visualisation for email archives in MBOX format, while the physical paper archive anchors the provenance chain. Springer
The British Library — one of the world's most sophisticated digital preservation institutions — is not treating paper as a legacy problem to be eliminated. It is treating the physical record as the provenance anchor against which digital surrogates are indexed. This is the hybrid model in practice, at the highest institutional level.
The DPC's four-point AI plan frames this with an explicit warning: if AI-generated content is not carefully managed, it will find its way into official and legal processes and form a completely erroneous historical record. There's a wall of content coming our way and how we handle it could have profound implications. The DPC does not propose a purely digital countermeasure. It proposes governance, standards, and — implicitly — the preservation of non-AI-generated physical reference points. Digital Preservation Coalition
Finding 4: Market Data Confirms Microfilm Is Not Declining — It Is Growing, Driven by the Exact Institutions That Understand Long-Term Preservation
Two independent market research reports confirm measurable demand growth for microfilm as an archival medium — sourced from the same institutions (government archives, national libraries, legal repositories) whose preservation mandates run 25–500 years.
The global market for microfilm and microfiche equipment and supplies was valued at USD 1.2 billion in 2023 and is expected to grow at a CAGR of 5.5% between 2025 and 2033, reaching USD 2.1 billion by 2033. This growth is driven by continued demand for long-term document storage solutions, particularly from government, libraries, archives, and other institutions requiring cost-effective and reliable methods of preserving large volumes of historical documents. The demand for microfilm equipment is particularly strong in government agencies, libraries, and archives that prioritise long-term document preservation. datahorizzonresearch
A second analyst report corroborates the direction: the microfilm equipment market is expected to grow from roughly USD 943 million in 2023 and reach about USD 1,342 million by 2032. North America shows stable demand, with the United States at a 3.1% CAGR supported by the ongoing modernisation of legal, government, and medical record repositories. Japan grows as institutions maintain large microfilm libraries and upgrade to hybrid reader-scanner systems for long-term preservation. Across all regions, demand is reinforced by the stability of microfilm as an archival medium. Future Market Insights
This growth is being driven by the same institutions responding to the digital preservation gap — not by organisations ignorant of digital alternatives, but by those who have evaluated both and are choosing hybrid.
Finding 5: NARA Still Mandates Silver Gelatin Microfilm for Permanent Federal Records — the Only Archivally Standardised Medium with a 500-Year Life Expectancy Rating
Despite mandating all-digital record transfers from agencies in 2024, NARA simultaneously maintains its federal regulation requiring that when records are microfilmed, agencies must use polyester-based silver gelatin type film that conforms to ISO 18901 for LE 500 film in all applications. "LE 500" is a formally standardised Life Expectancy rating of 500 years — assigned under an independent ISO standard. No digital format carries a comparable independently standardised life expectancy rating. This is not a coincidence or an oversight. It reflects the reality that no digital medium has been shown to last 500 years, and no digital format standard has survived unchanged for even 25. eCFR
Meanwhile, the Library of Congress explicitly states that preservation microfilming has the advantage of providing stable, long-term storage that is less prone to file corruption and loss of means to read a certain file type — and that it is an economical and proven option to preserve the intellectual content of collections on a large scale. The Library of Congress has maintained microfilm collections continuously since the late 1930s — the only independently verifiable proof of 85-year archival medium continuity under institutional custody that currently exists. No digital medium created in the 1980s, let alone the 1930s, is still readable today without multiple migration cycles. LOC
The Synthesis: Three Statements That the Evidence Supports and No Credible Third-Party Source Has Refuted
Statement 1: There is currently no validated, standardised digital solution for authenticating records created or managed by agentic AI systems over 25-year timeframes. The world's leading research consortium (InterPARES Trust AI, 2021–2026) is still investigating what the requirements should be. ASIS&T 2025 identifies the "evidential foundations" question as still open. The OAIS standard was revised in 2025 partly to address this gap. No published, ratified solution exists.
Statement 2: The archival science community's most credible institutions have already concluded that digital preservation cannot be achieved through technological means alone. This is not a commercial argument — it is the formally recorded position of the US Congressional commission on digital preservation (Principles for Digital Preservation), confirmed by the Digital Preservation Coalition, and enacted in practice by the British Library's hybrid analogue-digital archival programmes.
Statement 3: Microfilm's market is growing, its regulatory standing is unchanged, and its physical permanence properties remain unmatched by any digital medium at the 25–500 year horizon. Two independent analyst reports confirm a growing market. NARA's Code of Federal Regulations (36 CFR Part 1238) mandates ISO 18901 LE 500 silver gelatin film for permanent federal records. The Library of Congress holds 85 years of continuously readable microfilm collections — the only empirically proven long-term archival medium continuity that exists.
Selected Reference List
- Frontini, E. et al. (2022). "Trusted Data Forever: Is AI the Answer?" arXiv 2203.03712. InterPARES Trust AI / University of Macerata. Retrieved from: https://arxiv.org/pdf/2203.03712
- Jaillant, L. & Zhao, L. (2025). "Introduction: When data turns into archives." AI & Society, DOI: 10.1007/s00146-025-02374-y. Springer.
- Riter, M. (2025). "Representation and Authenticity in AI Generated, Curated, and Mediated Archives." Proceedings of ASIS&T 2025. Wiley Online Library. DOI: 10.1002/pra2.1498
- Duranti, L. et al. (2025). "From Traditional Archival Knowledge to Computer Vision and NLP." Nordic Archivaria, DOI: 10.18261/naf.32.1.2. InterPARES Trust AI findings.
- Digital Preservation Coalition (DPC). (2024). "DP and Artificial Intelligence — A Four Point Plan." DPC Blog. Retrieved from: https://www.dpconline.org/blog/dp-and-artificial-intelligence-a-4-point-plan
- ISO. (2025). ISO 14721:2025 — OAIS Reference Model, 3rd edition. Geneva: ISO. Retrieved from: https://www.iso.org/standard/87471.html
- NARA. (Current). 36 CFR Part 1238 — Microforms Records Management. US Code of Federal Regulations. Retrieved from: https://www.ecfr.gov/current/title-36/chapter-XII/subchapter-B/part-1238
- Library of Congress. (2025). "Preservation Microfilming — Conservation Treatment and Reformatting Options." Retrieved from: https://guides.loc.gov/preserving_newspaper/treatment
- Datahorizzon Research. (2024). Microfilm and Microfiche Equipment & Supplies Market — Global Market Size, Forecast 2024–2033. Retrieved from: https://www.datahorizzonresearch.com/microfilm-and-microfiche-equipment-and-supplies-market-45237
- Future Market Insights. (2025). Microfilm Readers and Scanners Market — Global Market Analysis Report 2035. Retrieved from: https://www.futuremarketinsights.com/reports/microfilm-readers-and-scanners-market
- Thibodeau, K. et al. (2004). Principles for Digital Preservation. arXiv cs/0411091. (US Congressional commission research basis.)