International Standards Bodies Explore Global Partnership for Artificial Intelligence Standardisation

ETSI has signed a letter of cooperation with Japan's Telecommunication Technology Committee (TTC) and Korea's Telecommunications Technology Association (TTA) to explore a new global partnership for AI standardisation. The announcement was issued alongside ETSI's unveiling of a new standard intended to reduce the production of harmful deepfakes. Separately, IEC, ISO and ITU issued the Seoul Statement at the International AI Standards Summit held 2–3 December in Seoul, Korea, committing the three bodies to advance AI standards for trustworthy AI systems.

What this does not do, as of publication: it does not change any legal obligation under the EU AI Act, and it does not create a harmonised standard citable in the Official Journal. It changes the direction of travel for the technical specifications that will eventually sit underneath those obligations.

Standards matter here because the EU AI Act delegates a substantial part of its regulatory detail to standardisation. Conformity for high-risk AI systems is intended to run through harmonised standards, which means the technical content of compliance is being written by standards development organisations rather than appearing in full in the legal text.

In Europe, that work sits with CEN and CENELEC Joint Technical Committee 21 (JTC 21), which was established on 1 June 2021 and is dedicated to developing European standards for artificial intelligence. According to CEN-CENELEC, JTC 21 unites over 300 experts from more than 20 countries working through five specialised working groups. Its stated objectives include addressing European market needs, supporting EU legislation and policies, developing harmonised standards for the AI Act, providing guidance to other technical committees, and adopting international standards or developing European ones where needed.

That last objective is the hinge. JTC 21 may adopt international standards rather than write European ones from scratch. Which means what ISO/IEC, ITU and — in the telecommunications domain — ETSI, TTC and TTA produce has a direct line into what eventually becomes a harmonised standard for artificial intelligence in the EU.

At the international level, the joint ISO/IEC committee SC 42 develops standards for responsible AI adoption using an ecosystem approach that considers regulatory, business, societal and ethical perspectives. IEC, ISO and ITU, through World Standards Cooperation, cover the entire AI ecosystem including terminology, governance, risk management, cybersecurity and ethical considerations.

Two distinct developments, from two different clusters of organisations.

ETSI signed a letter of cooperation with TTC (Japan) and TTA (Korea) to explore a new global partnership for standardisation in artificial intelligence. The ETSI announcement frames this as exploratory — a letter of cooperation, not a formal partnership project with a defined work programme and deliverable schedule. Practitioners should read it as an intent to align, not as a signed technical work plan.

The same ETSI communication cycle carried the announcement of a new ETSI standard to reduce the production of harmful deepfakes. The source material available confirms the existence and purpose of the standard but does not provide its deliverable number, its technical content, or its publication date. That detail is not verified here and should be taken from the ETSI deliverable register directly before anyone cites it in a technical file.

IEC, ISO and ITU issued the Seoul Statement at the International AI Standards Summit held 2–3 December in Seoul, Korea. It is a joint commitment to advance AI standards for trustworthy AI systems that benefit society and uphold fundamental rights.

The Seoul Statement sets out four commitments: incorporating socio-technical dimensions in standards development, addressing human rights, strengthening an inclusive multistakeholder community, and enhancing public-private collaboration on AI capacity building.

The ITU publication AI Standards for Global Impact (2025) summarises outcomes from the AI for Good Global Summit 2025, which united ISO, IEC and 53 UN partners to advance global standardisation across multiple AI sectors.

Harmonisation in the EU sense is a formal legal mechanism, and nothing announced here triggers it. A harmonised standard only confers presumption of conformity once it has been developed under a Commission standardisation request, assessed, and cited in the Official Journal of the European Union. The ETSI–TTC–TTA letter of cooperation is outside that chain. The Seoul Statement is outside that chain.

The chain that does matter runs: Commission standardisation request → CEN-CENELEC JTC 21 drafting (or adoption of an international standard) → Harmonised Standards Consultant assessment → citation in the Official Journal → presumption of conformity for the relevant AI Act requirement.

Where international alignment enters is at the second step. CEN-CENELEC states explicitly that JTC 21 may adopt international standards or develop European ones when needed. If ISO/IEC SC 42 deliverables and ETSI work converge with Japanese and Korean positions, the probability of adoption-rather-than-duplication rises. That is the mechanism through which a cooperation letter between telecommunications standards bodies eventually touches a CE marking file.

The European Commission maintains a dedicated page on standardisation of the AI Act, and the Rolling Plan for ICT Standardisation includes an artificial intelligence chapter for 2026. Both are the authoritative places to track whether international deliverables are being pulled into the European harmonised standards pipeline. This article does not have source material confirming the current citation status of any specific AI Act harmonised standard, and does not assert one.

Three markets are named directly in the cooperation: the EU, Japan and Korea. ETSI is a European Standardisation Organisation; TTC and TTA are the national telecommunications standards bodies of Japan and Korea respectively.

The product categories most exposed are those where AI functionality sits inside telecommunications and connected equipment — the domain where ETSI, TTC and TTA deliverables are routinely referenced. The deepfake standard additionally touches any product or service that generates, hosts, transmits or distributes synthetic media. The signal metadata for this development flags cybersecurity and IoT as the affected domains.

For manufacturers already running EU AI Act gap assessments, the practical impact is on standards selection strategy, not on current obligations. If a manufacturer is building an internal control framework against ISO/IEC SC 42 deliverables, convergence between European and Asian bodies reduces the risk of having to maintain parallel technical evidence for the same AI system across three regions. If a manufacturer is building against a region-specific specification, that bet gets riskier.

I will not quantify the impact. The sources provide no figures on the number of deliverables in scope, the timeline of the partnership, or adoption rates.

ETSI has announced a new standard aimed at reducing the production of harmful deepfakes. That is the extent of what the available official source material confirms.

What is not confirmed and must not be assumed: the deliverable reference number, whether it is a Technical Specification or a European Standard, whether it imposes provenance or content-credential requirements, whether it addresses detection, labelling, or generation-side controls, and whether it has any relationship to the transparency obligations for AI-generated content in the EU AI Act.

Anyone drafting an internal position on AI-generated content labelling standards requirements should pull the deliverable from the ETSI standards database before committing it to a compliance procedure. Treat any secondary summary of its content as unverified.

Nothing, as a matter of legal obligation. Neither a letter of cooperation nor a joint statement by standards bodies creates a requirement on manufacturers, importers or authorised representatives.

What changes is the planning assumption. Three specific adjustments are defensible on the available sources.

First, standards monitoring scope. If your AI standards watch list covers only CEN-CENELEC JTC 21 and ISO/IEC SC 42, it is now incomplete. ETSI deliverables — particularly in synthetic media and AI security — need to be in scope, and TTC and TTA outputs become relevant for Japanese and Korean market access planning.

Second, terminology alignment. IEC, ISO and ITU cover terminology across the AI ecosystem through World Standards Cooperation. Internal AI risk taxonomies that do not map to ISO/IEC SC 42 terminology will create friction during conformity assessment and during any notified body interaction under the AI Act.

Third, socio-technical and human rights documentation. The Seoul Statement commits IEC, ISO and ITU to incorporating socio-technical dimensions and addressing human rights in standards development. The practical read — flagged here as inference, not as a source-backed requirement — is that future deliverables from these bodies are more likely to require documented impact analysis beyond pure technical performance metrics. Teams that currently document only accuracy, robustness and cybersecurity should expect that evidence base to widen.

The confirmed dates in the source material are limited and should be treated as the only ones.

CEN-CENELEC JTC 21 was established on 1 June 2021. The International AI Standards Summit at which IEC, ISO and ITU issued the Seoul Statement was held 2–3 December in Seoul, Korea. The ITU publication AI Standards for Global Impact reflects outcomes from the AI for Good Global Summit 2025. The Rolling Plan for ICT Standardisation artificial intelligence chapter is published for 2026.

No date has been published for the conclusion of the exploratory phase of the ETSI–TTC–TTA partnership, and no deliverable schedule has been announced. There is no enforcement dimension to either development — standards bodies do not enforce. Enforcement attaches only when a harmonised standard is cited in the Official Journal and a market surveillance authority assesses a product against the underlying legal requirement.

The structural difference is that the EU has a binding legal instrument — the AI Act — that delegates technical detail to harmonised standards, creating a direct legal consequence for standards content. The available sources do not describe an equivalent legal mechanism in Japan or Korea, and this article does not assert one.

What the sources do establish is institutional: ETSI, TTC and TTA are the three signatories to the cooperation letter, representing Europe, Japan and Korea in telecommunications standardisation. Korea additionally hosted the International AI Standards Summit at which the Seoul Statement was issued by IEC, ISO and ITU.

For a compliance team, the operating assumption should be that European AI standards carry regulatory weight through the AI Act's harmonised standards mechanism, while international and Asian deliverables currently carry commercial and interoperability weight — with the cooperation letter signalling that the gap between those two categories is intended to narrow.

Start with the deepfake standard. Pull the ETSI deliverable directly from the ETSI standards database rather than relying on the press announcement. Determine its reference number, its status, and its normative scope. If your product generates, modifies, hosts or distributes synthetic media — including voice cloning, face swapping, or generative image features embedded in consumer devices — assign an engineer to read it in full and produce a one-page applicability note. Do not wait for a legal trigger; the technical lead time on provenance and content-credential features is measured in release cycles, not weeks.

Then fix your standards watch list. Add ETSI AI and synthetic media deliverables, ISO/IEC JTC 1/SC 42 work items, ITU-T AI-related recommendations, and CEN-CENELEC JTC 21 work programme updates. If you sell into Japan or Korea, add TTC and TTA. Assign a named owner — not a shared inbox — and set a monthly review.

Run a terminology mapping exercise against ISO/IEC SC 42 vocabulary. Take your internal AI risk register, your model documentation templates and your technical file structure, and map every term to the SC 42 equivalent. Where there is no mapping, that is a finding. This is cheap to do now and expensive to retrofit once a notified body is reading your file.

For AI Act high-risk classification work, keep the gap assessment anchored to the legal text and the Commission's standardisation page, not to draft standards. Draft deliverables change. Build your quality management system so that the standard reference is a configurable input, not hard-coded into procedures — when JTC 21 adopts an international standard rather than publishing a European original, you want to swap the reference without rewriting the procedure.

Finally, document the socio-technical dimension now. The Seoul Statement's four commitments include socio-technical dimensions and human rights. Whether or not that converts into normative requirements, building the evidence trail — intended use analysis, foreseeable misuse, affected population analysis — costs little during development and is close to impossible to reconstruct retrospectively.

Three specific triggers. First, the publication of the ETSI deepfake deliverable in the ETSI standards database with a full technical specification — that is the item with the nearest-term engineering impact. Second, any conversion of the ETSI–TTC–TTA letter of cooperation into a formal partnership project with a published work programme; until that happens, the partnership is exploratory and should be tracked, not planned around. Third, the artificial intelligence chapter of the Rolling Plan for ICT Standardisation 2026 and the European Commission's AI Act standardisation page, which together are the authoritative indicators of whether international AI standards are being drawn into the European harmonised standards pipeline.