Conference October 2026 📍 Portland, Oregon, USA

PDF Technical Conference 2026 — Portland: PDF 2.0 Adoption and AI-Driven Document Workflows

The PDF Technical Conference 2026, held in Portland, Oregon in October 2026, brought together PDF implementors, specification authors, and enterprise adopters to discuss the state of PDF 2.0 adoption, AI-driven document processing, PDF/UA-2 accessibility, and the evolving role of PDF in an AI-first document landscape.

Official event page →

Key takeaways

PDF 2.0 creator tool support is mature, but the enterprise installed base remains predominantly PDF 1.4–1.7; the transition is gradual.
AI document processing has made PDF text extraction quality a first-order business concern — low-quality PDFs produce low-quality AI outputs.
Tagged PDF investment serves both accessibility and AI extraction quality — a convergence driving accessibility investment beyond regulatory compliance.
PDF/UA-2 validator and authoring tool support is maturing; AI-assisted remediation is emerging as a scalable approach for large document libraries.
AI-generated PDFs are raising new questions about provenance, authenticity, and the applicability of digital signatures to AI-authored content.

The PDF Technical Conference 2026, held in Portland, Oregon in October 2026, was organised by the PDF Association. The 2026 conference took place against a backdrop of significant shifts in the document technology landscape: PDF 2.0 (ISO 32000-2:2020) had been published six years prior and was now at a point where adoption questions were giving way to implementation maturity questions. AI-driven document processing had moved firmly into enterprise production. PDF/UA-2 (ISO 14289-2:2024) was two years old. The Portland conference brought together PDF implementors, enterprise document architects, accessibility specialists, and standards committee members to assess where PDF stood and where it was heading. Portland was selected as the venue to complement the European PDF Days series with a North American technical conference, reflecting the geographic distribution of the PDF standards working groups and the enterprise document management market.

PDF 2.0 adoption: six years on

ISO 32000-2:2020 (PDF 2.0) had been published in 2020, with an earlier edition in 2017. By October 2026, the adoption picture was a study in contrasts: Creator tool support — Major PDF creation tools and libraries (Adobe Acrobat, iText, PDFlib, Apache PDFBox, pdfium, PDF.js) all had meaningful PDF 2.0 support. PDF 2.0 features including PDF 2.0 digital signatures, enhanced metadata, and improved tagged PDF structures were being produced in new workflows. Adobe's Creative Suite had PDF 2.0 output as a selectable option in export dialogs. Enterprise document libraries — The installed base of PDFs in enterprise document management systems remained predominantly PDF 1.4–1.7. Organisations were not retroactively converting existing archives to PDF 2.0 — the cost-benefit was not there. New documents being generated were increasingly PDF 2.0, but this transition would take years to be reflected in the overall document corpus. Viewer support and the long tail — PDF 2.0 documents with features beyond the 1.7 subset required up-to-date viewers for full functionality. Older PDF viewers (embedded in legacy enterprise applications, regulatory submission portals, government document management systems) still accepted PDF 2.0 documents but might silently ignore features. This created a pragmatic pressure to limit use of PDF 2.0-specific features to contexts where viewer compatibility could be controlled. PDF 2.0 errata and clarifications — The ISO committee had published several corrigenda and Technical Specifications addressing ambiguities in ISO 32000-2. The Portland conference included sessions on how implementors had encountered and resolved ambiguities in practice.

AI-driven document workflows: the PDF quality imperative

By 2026, AI document processing was a mainstream enterprise capability. Large language models and multimodal AI systems were being used for contract analysis, regulatory document review, scientific literature processing, invoice extraction, and automated document classification. The Portland 2026 conference had a substantial programme on the intersection of AI and PDF: PDF as AI input format — The quality of AI outputs on PDF documents is a direct function of the quality of text extraction from those PDFs. Embedded text with complete /ToUnicode mappings gives accurate extraction; scanned PDFs require OCR; PDFs with glyph-level encoding issues produce garbled text. Organisations running AI on document corpora were discovering PDF quality problems at scale that had never been apparent when humans were the consumers. Multimodal AI and PDF rendering — Multimodal models (capable of processing both text and images) opened new approaches to PDF processing that bypassed the text extraction problem: render each page as an image and process visually. This approach had significant cost and quality trade-offs compared to text extraction, but was increasingly viable for documents where text extraction quality was too low to be reliable. Sessions discussed when to use text extraction vs. visual processing approaches. AI-generated PDFs — A growing category of PDFs were now being generated by AI systems: AI-written reports rendered to PDF, AI-synthesised documents, AI-assembled document collections. Questions about metadata provenance, content authenticity, and the applicability of digital signatures to AI-generated content were new territory for the standards community. PDF/UA and AI accessibility — Structured, tagged PDF was doubly valuable in an AI era: it served both accessibility requirements and AI structure extraction. The investment in PDF/UA-compliant tagging produced documents that were better for both screen readers and language models. This convergence was being cited as a business case for accessibility investment beyond regulatory compliance.

PDF/UA-2 implementation: the 2026 status

PDF/UA-2 (ISO 14289-2:2024) was published in 2024. By the time of the Portland 2026 conference, implementation was two years in: Validator maturity — veraPDF and PAC had mature PDF/UA-2 validation, with the inter-validator discrepancies of 2025 largely resolved through coordination between the validator developers. The PDF Association's Matterhorn Protocol had been updated to cover PDF/UA-2 requirements, providing a machine-testable interpretation of the standard. Authoring tool progress — Microsoft Word, Adobe InDesign, and LibreOffice had improved PDF/UA-2 export support. The new structure elements introduced in PDF/UA-2 (/Aside, /FENote, /Sub, /Em) were being produced by at least some tools. Auto-tagging tools using AI to generate structure trees for existing untagged PDFs had improved significantly — AI-assisted remediation was becoming a viable option for large document libraries. MathML in scientific publishing — The academic and scientific publishing sector had made meaningful progress on MathML embedding for mathematical content. LaTeX-to-PDF pipelines with MathML output were in production at several major publishers. This represented real accessibility improvement for the technically dense content types where PDF/UA had previously been most difficult to achieve. Procurement and regulatory drivers — European Accessibility Act requirements (effective 2025) and updated US Section 508 guidance were creating regulatory pressure for PDF/UA compliance in public-sector and commercial web content. The Portland conference included sessions on the compliance landscape and what "sufficient PDF/UA compliance" looked like in regulatory practice.

Frequently asked questions

What are the key new features in PDF 2.0 compared to PDF 1.7?

PDF 2.0 (ISO 32000-2) introduces several significant additions over PDF 1.7/ISO 32000-1: (1) New page-level metadata — page-level XMP metadata streams, improving document information granularity. (2) Digital signature improvements — PAdES compatibility enhancements and clearer requirements for long-term validation (LTV) signatures. (3) New structure types — additional tagged PDF structure elements for improved accessibility and document semantics. (4) Black point compensation — standardised treatment for colour space black point compensation. (5) Encryption improvements — clarifications and improvements to the encryption model. (6) Removal of deprecated features — several obsolete features from early PDF versions were formally removed from the normative specification. (7) Many clarifications and ambiguity resolutions — PDF 2.0 addressed hundreds of unclear or contradictory requirements in ISO 32000-1 based on implementation experience.

How does AI processing affect PDF quality requirements?

AI processing of PDFs amplifies the consequences of PDF quality defects that were previously inconsequential. When humans read PDFs, they can compensate for extraction artefacts — a garbled word is re-read from the visual rendering, context fills gaps. AI language models processing extracted text have no such fallback: garbled text from incomplete /ToUnicode mappings, ligature characters without Unicode mappings, or OCR errors in scanned PDFs produce inputs that cause AI models to generate inaccurate outputs — and to do so confidently, without signalling that the input was corrupted. Organisations deploying AI on PDF corpora need to assess document quality (text extractability, OCR quality for scanned files) before relying on AI outputs.

What is AI-assisted PDF remediation and how does it work?

AI-assisted PDF remediation is the use of machine learning models to automatically generate or improve accessibility structure (tagging) in existing untagged or poorly tagged PDFs. The general approach: a computer vision model analyses the rendered page image to identify structural elements (headings, paragraphs, tables, figures, lists); a language model or rule-based system generates the corresponding PDF tag structure tree; the tags are written back into the PDF using a PDF editing library. The quality of AI remediation varies significantly with document complexity. For cleanly structured single-column text documents, AI remediation can approach manual quality. For complex multi-column layouts, tables with merged cells, and documents with heavy graphical design, AI remediation typically requires human review and correction. Tools offering AI-assisted remediation include commercial products from Axes4, CommonLook, and newer AI-native tools, as well as API services from major cloud providers.

Try it in your browser

The PDF tools and standards discussed at this event are available free — no account needed.

Edit PDF Sign PDF Protect PDF Open editor