PDF/UA-2 (ISO 14289-2:2024) had been published for approximately one year when PDF Days Europe 2025 convened. Sessions reported on where the ecosystem actually stood: Validator support — The PAC (PDF Accessibility Checker) had shipped PDF/UA-2 validation. veraPDF had PDF/UA-2 support in production releases. However, both validators were still identifying edge cases and issuing updates — first-year validator implementations always do. Organisations that had started targeting PDF/UA-2 were finding that passing PAC did not guarantee passing veraPDF and vice versa on some specific requirements, indicating that validator consensus was not yet complete. Creation tool support — Microsoft Word's PDF export, Adobe InDesign, and PDF creation libraries had varying degrees of PDF/UA-2-compatible output. The new structure elements (/Aside, /FENote) were not yet produced by most authoring tools, meaning that organisations wanting to use them needed manual post-processing in Acrobat Pro or specialist tagging tools. The no-/B-level consequence — The elimination of the /B (basic) conformance level in PDF/UA-2 was having concrete effects. Organisations that had achieved PDF/UA-1/B compliance and considered themselves "done" were discovering they needed to go further to meet the full PDF/UA-2 standard. For large document libraries (government agencies, financial services firms), this represented significant remediation scope. MathML progress — Academic and scientific publishers were beginning to adopt MathML embedding for mathematical content, with the LaTeX-to-PDF pipeline tooling being adapted. This was one of the areas where PDF/UA-2 was opening genuinely new capability rather than tightening existing requirements.

By 2025, AI document processing had moved firmly into production. Large language models deployed on enterprise document corpora — for contract analysis, regulatory filing review, scientific literature mining, and invoice processing — were producing real-world evidence about PDF quality at scale. Sessions at PDF Days Europe 2025 reported on what those deployments had revealed: The /ToUnicode quality problem at scale — AI text extraction quality correlates strongly with the quality of /ToUnicode mappings in embedded fonts (see Text Layer). In enterprise document corpora, a meaningful percentage of PDFs — particularly older PDFs from legacy publishing systems — have missing or incomplete /ToUnicode mappings. AI systems processing these PDFs produce corrupted text extraction, and downstream AI outputs on corrupted text are unreliable regardless of model quality. Organisations deploying AI on document corpora were discovering they first needed to assess and repair the PDF quality of their archives. Tagged PDFs and AI structure understanding — Organisations that had invested in Tagged PDF workflows for accessibility were reporting a secondary benefit: AI systems processing tagged PDFs had better section and hierarchy understanding, improving summary and extraction quality. The accessibility investment was paying dividends in AI performance. Scanned documents and OCR quality — Scanned PDFs in AI pipelines require OCR pre-processing. Modern OCR (including cloud OCR APIs from AWS, Google, and Azure) achieves high accuracy on clean printed documents, but accuracy on historical documents, handwritten annotations, degraded scans, and non-Latin scripts remains variable. AI outputs on low-accuracy OCR text are confidently wrong — a significant risk in legal and medical document contexts. PDF/A-3 as the AI-friendly archiving format — The pattern of embedding machine-readable structured data alongside human-readable PDF/A-3 content (as in ZUGFeRD e-invoicing) was being extended to other document types: structured clinical data in healthcare records, contract metadata alongside contract PDFs, and research data alongside academic papers. This pattern avoids the OCR/extraction quality problem entirely for the structured data.

NIST's first post-quantum cryptography standards — ML-KEM (FIPS 203) and ML-DSA (FIPS 204) — were published in August 2024. For PDF digital signatures, which currently use RSA and ECDSA algorithms that a cryptographically relevant quantum computer could break, this publication moved the post-quantum transition from a theoretical concern to a concrete planning requirement. Sessions at PDF Days Europe 2025 addressed what this means in practice for PDF signing workflows: Timeline uncertainty — The consensus estimate for a cryptographically relevant quantum computer remains in the 10–20 year range, with significant uncertainty. However, "harvest now, decrypt later" attacks — where adversaries collect signed documents today intending to break signatures when quantum hardware is available — make the timeline for sensitive long-lived documents effectively immediate. PDF specification updates needed — The current PDF specification's encryption dictionary does not include identifiers for NIST post-quantum algorithms. Updates to ISO 32000 and the ETSI PAdES profiles will be needed before post-quantum signatures can be formally specified in PDF. The PDF Association's Technical Working Group was tracking the required changes. Hybrid signatures as the bridge — The practical near-term approach for high-value documents is hybrid signing: creating signatures with both a classical algorithm (ECDSA-256) and a post-quantum algorithm (ML-DSA). If the classical algorithm is broken in the future, the post-quantum signature still holds. If the post-quantum algorithm has undiscovered weaknesses, the classical signature still holds. This approach is used in TLS 1.3 post-quantum key exchange already. Who should act now — Documents that need to remain cryptographically verifiable for 20+ years — long-term contracts, land registry documents, archival medical records, government records — are the priority. Standard business documents with 3–7 year retention do not yet require post-quantum signatures.

The EU regulatory context had changed significantly in the year since PDF Days Europe 2024: European Accessibility Act enforcement started June 2025 — The European Accessibility Act (Directive 2019/882) enforcement began across EU member states in June 2025. National market surveillance authorities were beginning the complaint-handling and enforcement processes. Sessions reported on early enforcement patterns: which jurisdictions were most active, what complaint categories were appearing, and what remediation commitments regulators were accepting. B2B e-invoicing mandates accelerating — EU member state legislation requiring electronic invoicing for business-to-business transactions was advancing rapidly. France had mandated B2B e-invoicing for large businesses. Germany, Italy, Poland, and others were at various stages of legislation or implementation. The PDF/A-3 based ZUGFeRD / Factur-X format was the dominant technical response, and sessions reviewed the implementation challenges at scale.