On this page
The calendarTrend 1: AI rules reach the translation workflowTrend 2: accessibility becomes a localization requirementTrend 3: LLM drafts, human sign-offTrend 4: translation moves into the release pipelineWhat to do this quarterIn briefMost articles on translation and localization trends are lists of predictions with no dates and no sources. This one is the opposite: a short list of changes you can check, each tied to a rule, a standard or vendor documentation, and what each means for a company that publishes in Korean. The predictions can wait. These are already on the calendar.
The post was first published in 2025 and has been updated to reflect where those rules stand now. Where something is still being worked out, it says so. For the strategy side of the same question, see what drives localization ROI.
The calendar
| Date | What applies | Why localization teams care |
|---|---|---|
| 2 February 2025 | EU AI Act: general provisions, including AI literacy, and prohibitions | Staff who use AI translation tools at work are inside the AI literacy duty's reach |
| 28 June 2025 | European Accessibility Act enforceable in Member States | Covered e-commerce, banking and e-book services must be accessible in every language version |
| 2 August 2025 | EU AI Act: rules for general-purpose AI models | Obligations fall on model providers; buyers get more documentation to read |
| 22 January 2026 | Korea's AI Basic Act takes effect | Notice and labeling duties for generative AI operators serving Korea |
| 2 August 2026 | EU AI Act: transparency rules (Article 50) | Disclosure duties for certain AI interactions and AI-generated content |
Dates for the EU AI Act are from the European Commission's AI Act Service Desk, which notes that its timeline reflects amendments made by the Digital Omnibus on AI. Check that page before relying on any later deadline.
Trend 1: AI rules reach the translation workflow
Neither the EU AI Act nor Korea's AI Basic Act is a translation law, but both touch teams that use AI to produce content in other languages.
What to do: keep a simple register of where AI produces or drafts text in your Korean workflow, who reviews it, and what is published without review. That register is what you will need for any of these questions. For a deeper look at the EU side, see an EU AI Act–ready localization playbook.
Trend 2: accessibility becomes a localization requirement
The European Accessibility Act has been enforceable since 28 June 2025. It covers, among others, e-commerce services, consumer banking services, e-books and e-reading software, and passenger transport websites and apps. Accessibility does not stop at the English version. Alt text, form labels, error messages and screen-reader text all need translating, and they are exactly the strings most often left out of a localization export because nobody sees them on screen.
Korea is not an EU market, but companies that serve both usually run one product. If your Korean version is built from the same code, add accessibility strings to the localization scope and test the Korean build with a screen reader. Text that is hidden visually is still read aloud.
Trend 3: LLM drafts, human sign-off
Machine translation has moved from separate engines to large language models built into translation tools, content systems and office software. The practical change is not that drafts got better overnight. It is that drafts are now everywhere, including in tools nobody approved for translation.
Two controls separate a working setup from a risky one. First, constrain the draft: glossaries and translation memory should steer the output. Google's Cloud Translation Advanced documentation, for example, describes glossaries that fix the translation of specific terms and notes that its newer LLM-based model supports a contextual glossary option. Second, define the human step. ISO 18587 sets requirements for full human post-editing of machine output and post-editor competence; ISO lists it as due for revision, so watch for an update. For the review side, see neural MT with human QA.
Trend 4: translation moves into the release pipeline
Software and e-commerce teams increasingly connect their code and content systems directly to translation management systems, so Korean ships with each release instead of weeks after. The shift changes what buyers need from a vendor: smaller batches, faster turnaround, reviewers who work inside the tools. For the setup, see translation software categories for buyers.
What to do this quarter
- Map where AI drafts or produces Korean text, and who approves it.
- Give everyone who uses those tools a short briefing on their failure modes.
- Ask counsel which of your Korean-facing services fall under the AI Basic Act's notice and labeling duties.
- Add alt text, form labels and screen-reader strings to the localization scope.
- Load your glossary into every MT or LLM tool in use, and test it on real content.
- Review the plan against the budget; see planning a yearly translation budget.
In brief
Does the EU AI Act regulate machine translation?
It does not single out translation. But its AI literacy duty already applies to organisations deploying AI systems, and its transparency rules for certain AI-generated content apply from August 2026. If your localization uses AI, check how both affect your workflow and published content.
Does Korea's AI Basic Act require labeling machine-translated text?
The Act requires generative AI operators to notify users and label generated content, with specific notice for synthetic audio, images and video that are hard to tell from real. How this applies to translated text is a question for counsel, especially while enforcement details are still being set.
Are LLMs replacing translation memory?
Not in well-run workflows. LLM output is typically constrained by the memory and glossary, not the other way round. The memory remains the record of approved translations.
Checking your Korean workflow against the new rules?
We review AI-drafted Korean and help you set up glossaries and sign-off steps that hold up. See our services →