Industries / Government, education and community / Nonprofits / Standards bodies
Digital standards
This work turns standards from documents for people to read into data that software can search, check and reuse.
4
Milestones achieved, from 2011 to 2024.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2011
ISO publishes in XML
At the end of 2011 ISO rebuilt its publishing system and, with Mulberry Technologies, a US markup firm, made the ISO Standards Tag Set (ISOSTS): a set of XML tags for standards adapted from JATS, the tag set that journals use for articles. XML labels each part of a document, such as a clause, a table or a defined term, so that software can find it. By the account of NISO, the US body for information standards, ISOSTS then ran "in full production" with few changes and was adopted by BSI (UK), SIS (Sweden), NEN (Netherlands) and Standards Australia.
LessonMarking structure did not make the content usable by machines: in 2025 CEN still wrote that standards users "invest significant time and effort to integrate content from PDF standards" into their own systems.
Credit
ISO; Mulberry Technologies
2017
One tag set for publishers
US standards writers held back from ISOSTS because it was not itself a standard, NISO says. A NISO committee drafted the Standards Tag Suite (STS) in 2016 and 2017, tied to JATS so that the two change together, and it was published as ANSI/NISO Z39.102-2017; version 1.2 followed in November 2022. It includes TBX-ISO-TML, a markup for terms and their definitions, and its schemas and user documentation are free to download.
Credit
NISO and its STS committees
2019
A ladder to machine use
ISO began work in 2018 and in 2019 proposed SMART standards ("Standards Machine Applicable, Readable and Transferable"), graded in five levels, by the account of researchers at the China National Institute of Standardization (CNIS). A joint ISO and IEC leaflet defines them: level 0 is paper, level 1 PDF and level 2 XML, where "the machine can understand the structure"; at level 3 "the machine can identify content from the standard", and at level 4 it "can take that content, interpret and use it". ISO started its SMART project in May 2021 and tested 10 pilot projects in 2023, and with IEC, CEN and CENELEC released a "Powered by SMART" demonstration, CNIS reports. CEN, which credits IEC with developing the level model, says ISO, CEN and CENELEC adopted it, and CEN and CENELEC aim for level 3. China's standards administration (SAC) issued its own guideline for building standards on standards digitalization on 3 April 2025.
webdesk.jsa.or.jpscience.cspress.com.cnexperts.cen.eusac.gov.cn
Credit
ISO and IEC, through their joint SMART programme
2024
Standards drafted in the browser
On 6 August 2024 IEC published IEC 60512-28-100:2024, a set of signal tests for electrical connectors up to 2,000 MHz, the first international standard developed entirely on the Online Standards Development (OSD) platform, a shared web drafting tool of ISO, IEC, CEN and CENELEC, by CNIS's account. The same account says OSD had handled 90% of ISO deliverables since June 2023, and that under a 2024 resolution of ISO's Technical Management Board OSD became the default from January 2025 for eligible new and revised ISO standards. CEN says it delivered its own SMART drafting process, from first draft to public enquiry, in January 2025, but keeps the number of projects low "as manual work is required".
Credit
IEC and ISO; CEN and CENELEC for their SMART drafting tool
Watch for
Answers cited to the clause
Accuris, which sells online access to standards from many publishers, launched an AI Assistant in its Engineering Workbench and Accuris Thread products that it says gives "clause-level answers" from "licensed standards and technical content", is "grounded entirely in publisher-authorized content", and "returns answers with precise citations to the underlying clauses". It gives no accuracy figures.
Achieved whenAn outside party scores the assistant on a published set of questions with answers and clause numbers checked by engineers, and reports both wrong answers and wrong citations.
Claimed by
Accuris, April 2026
DIN bets on data
DIN Media, the publishing arm of DIN, Germany's standards body, is building an AI assistant on Google's Gemini model that is "intended to guide users through cascading questions" to an answer with "a reference to the relevant clauses", and is looking to "granular data formats such as XML or ReqIF", the latter a format for exchanging requirements between engineering tools. Its head of partner management says many companies "wish to feed standards into their internal, AI-based systems".
Achieved whenDIN releases the assistant or the ReqIF data to paying customers and a user publishes a test of either on real standards.
Claimed by
DIN Media (Marion Winkenbach and Sabrina Butters), in KANBrief issue 2/26, 2026
Open
01
Tag every provision
An open set of NASA technical standards, which as works of the US government carry no copyright, converted to NISO STS XML, with every provision tagged by type (requirement, recommendation, permission or statement) and by its parts (who must act, under what condition, and any value with its unit). This is the level-3 markup that the SMART plans describe. Add an AI tagger and score it against people.
The briefCredit · unclaimed02
What changed between editions
An open test set of pairs of editions of public standards, each pair with a list, made and checked by two engineers, of every change to a requirement: added, removed, made stricter or looser, or a recommendation turned into a requirement. Add a script that scores any comparison tool on finding them, with an ordinary text comparison as the baseline.
The briefCredit · unclaimed
Open 01 · the brief
Tag every provision
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open set of NASA technical standards, which as works of the US government carry no copyright, converted to NISO STS XML, with every provision tagged by type (requirement, recommendation, permission or statement) and by its parts (who must act, under what condition, and any value with its unit). This is the level-3 markup that the SMART plans describe. Add an AI tagger and score it against people.
Why it is still open
CEN marks up provisions by hand in its SMART pilots and keeps their number low "as manual work is required"; ISO's SMART work had tested 10 pilots by 2023. GB/T-Bench, an August 2026 benchmark, tests AI models on reviewing drafts of Chinese national standards, including their use of "shall", "should" and "may", by planting errors, not on turning standards into data. The nearest labelled corpus, LexDeMod (2022), marks obligations and permissions in English contracts, a different kind of text. No open collection of standards tagged at level 3 was found.
experts.cen.euarxiv.orgscience.cspress.com.cnarxiv.orglaw.cornell.edu
Built with what exists
Standards downloaded free from NASA's Technical Standards System; NASA-STD-7009B, reformatted in 2024, numbers all its requirements and lists them in a requirements matrix in its Appendix A, which serves as a ready answer key; the NISO STS 1.2 schemas; Label Studio for human tagging; a model that scores answers from a fixed list, such as Google's Gemini Decision 1, to pick each provision's type (see the Note "A thousand questions, one pass"); a long-context model such as Claude to fill in the parts.
First step
Convert NASA-STD-7009B to STS XML and check whether an AI tagger finds each requirement numbered in its Appendix A.
Done looks like
20 NASA standards as valid STS XML; at least 5,000 provisions tagged by two people, with their agreement reported; the AI tagger finding every requirement listed in the requirements matrix of each standard that has one, with precision and recall for requirements of at least 0.95 against the human tags; all files under an open licence.
Open 02 · the brief
What changed between editions
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open test set of pairs of editions of public standards, each pair with a list, made and checked by two engineers, of every change to a requirement: added, removed, made stricter or looser, or a recommendation turned into a requirement. Add a script that scores any comparison tool on finding them, with an ordinary text comparison as the baseline.
Why it is still open
Publishers and vendors sell comparisons. ASTM sells "redline" versions of its standards that mark, with struck-through text, what was removed from the previous edition. Accuris says its Change Impact Analysis replaces a cycle in which teams "download editions, extract changes by hand, rebuild the spreadsheet", but it publishes no accuracy figures, and no public test of change detection on standards was found. Standards' own change logs give only highlights: the history log of NASA-STD-7009B lists "key changes" such as "promoting recommendations to requirements", so a complete list still has to be made by hand.
Built with what exists
Current and historical editions on NASA's standards site, such as NASA-STD-7009, 7009A and 7009B; Python's difflib for the text baseline; general AI models such as Claude and Gemini for comparison by meaning; a spreadsheet for the engineers' answer key.
First step
List every requirement change between NASA-STD-7009A with Change 1 and NASA-STD-7009B, and compare the list with the summary in 7009B's history log.
Done looks like
30 edition pairs with at least 1,000 labelled requirement changes, under an open licence, and a table giving each tool's share of real changes found and share of reported changes that are real.