AI-Assisted Translation and ISO 17100: 3 Powerful Compliance Lessons for Modern LSPs

Artificial intelligence is changing the translation industry at remarkable speed. What once seemed like a distant technological possibility has now become part of everyday language operations. Translation service providers are using AI tools for terminology support, quality checks, drafting assistance, consistency control, project preparation, and, in some cases, the generation of translated content itself.

This development raises an important compliance question: Can an AI-assisted workflow comply with ISO 17100?

The answer depends on the exact workflow. AI is not automatically incompatible with ISO 17100. However, a workflow where the actual translation is mainly generated by AI and then checked by a human may not fit comfortably under ISO 17100 alone.

ISO 17100 is built around a human-led translation process, with defined roles, competences, resources, project management controls, and procedures for delivering quality translation services. The standard focuses on the translation service itself, not simply the final appearance of the delivered text.

That distinction is essential. If a qualified human translator produces the translation and a qualified human reviser checks it, the workflow aligns with the traditional ISO 17100 model. But if AI or machine translation produces the first target-language version and a human edits that output, the process begins to move toward post-editing rather than standard human translation and revision.

For language service providers, the key is honesty in workflow definition. AI may support the process, but it should not quietly replace the required human translation role where ISO 17100 compliance is being claimed.

AI-Assisted Translation and ISO 17100

The relationship between AI-assisted translation and ISO 17100 must be understood carefully. Many organizations now use the term “AI-assisted” very broadly. In some cases, it means that a human translator uses AI for support, such as terminology research or quality checks. In other cases, it means that AI produces the first full translation and a human reviews or edits it afterward.

These are very different workflows.

In an ISO 17100 environment, the traditional translation process depends on qualified human professionals. The translator produces the translation. The reviser checks the translation against the source text. The project manager ensures that client requirements, resources, timelines, and quality controls are properly managed.

AI can be present in this environment, but its role must be controlled and clearly understood.

For example, an AI tool may help a translator compare terminology options, identify inconsistent phrasing, check formatting, or highlight possible quality issues. In this type of workflow, the qualified human translator remains responsible for creating the translation. The qualified human reviser remains responsible for checking it.

That is very different from a workflow where AI generates the translation and a human simply corrects the output.

The important question is not whether AI exists somewhere in the process. The important question is whether the required human roles under ISO 17100 remain real, active, qualified, and demonstrable.

If the human translator’s role is reduced to approving machine-generated output, then the service may no longer reflect the traditional ISO 17100 model. It may instead fall closer to post-editing, where ISO 18587 becomes more appropriate.

This is why LSPs should avoid unclear claims. Phrases such as “ISO 17100 AI translation” or “AI-powered ISO translation” can create confusion unless the workflow is fully explained.

A better approach is to describe the service accurately and transparently. For example, an LSP may state that it provides human translation and revision under ISO 17100, machine translation post-editing under ISO 18587, or AI-supported human translation where qualified human roles are maintained.

Clear language supports trust. It also helps clients understand what they are buying.

Understanding the Role of Human Translation

ISO 17100 is not simply about producing a final text that reads well. It is about the process used to deliver that text. The standard focuses on controlled translation services, qualified people, clear project management, client requirements, revision, and the ability to demonstrate conformity through documented evidence.

In a traditional ISO 17100 workflow, the translator is responsible for producing the translation. The reviser then examines the target-language content against the source-language content to identify issues related to meaning, accuracy, terminology, grammar, style, and suitability for purpose.

This workflow depends on human competence at each critical stage.

The translator must understand the source text, the target language, the subject matter, the client’s requirements, and the intended purpose of the translation. The reviser must then assess whether the target text accurately reflects the source and meets the required quality expectations.

This is not a minor administrative step. It is a core part of the quality process.

AI tools can still have a place in such an environment. A translator may use AI to research terminology, compare phrasing options, identify inconsistencies, support formatting, or speed up non-core administrative tasks.

In that case, AI is acting as an assistant, not as the main producer of the translation.

The compliance concern becomes stronger when the AI system creates the main translation output and the human role is limited to checking, correcting, or approving it afterward. That is no longer the same operational model as human translation followed by human revision. It is closer to machine translation post-editing.

This is why language service providers should avoid vague service descriptions unless the workflow is clearly explained. Clients, auditors, and certification bodies need to understand who produced the translation, who revised it, what technology was used, and how quality was controlled.

Human responsibility must be visible. The workflow must show that qualified professionals were not merely present in name, but actively responsible for the work.

When ISO 18587 Becomes the Better Fit

ISO 18587 is the more natural standard for workflows based on machine translation output followed by human post-editing.

When AI or machine translation generates the first translated draft, and a human professional then edits and corrects that output, the service should normally be treated as post-editing. In such a case, ISO 18587 gives a more appropriate compliance framework than ISO 17100 alone.

The difference is not just a matter of wording. “Revision” and “post-editing” are different professional activities.

Revision usually means checking a human translation.

Post-editing means editing machine-generated translation output.

These two processes may involve similar language skills, but they are not identical. They have different starting points, different risks, and different professional expectations.

A reviser evaluates a translation produced by a human translator. A post-editor evaluates and improves output produced by a machine translation or AI system.

This distinction is important because machine-generated output can create unique types of errors. It may be fluent but inaccurate. It may preserve the structure of the source language in unnatural ways. It may mistranslate terminology. It may omit details. It may use inconsistent phrasing. It may create a sentence that looks convincing but changes the meaning.

Post-editors must be trained to detect these specific risks.

A language service provider can offer both ISO 17100 and ISO 18587 services, but it should label them correctly. Human translation with revision can be managed under ISO 17100. Machine translation post-editing can be managed under ISO 18587.

Hybrid workflows may require careful assessment, depending on whether the human or the AI system is responsible for creating the actual translation.

The safest approach is not to force all AI-related work into ISO 17100. The safer and more professional approach is to classify each service according to the real workflow used.

The Risk of Light Human Review

One of the biggest compliance risks in the current market is the use of AI-generated translation followed by only a light human review. This may appear efficient, especially when deadlines are tight or budgets are limited. However, it can create problems if the service is presented as ISO 17100-compliant human translation.

A quick review of AI output is not the same as full translation revision. A reviewer may miss subtle meaning shifts, omissions, terminology inconsistencies, register problems, cultural issues, or legal and technical inaccuracies.

AI-generated language can often sound fluent while still being wrong. That makes professional human responsibility even more important.

This is one of the greatest challenges with modern AI translation. Poor machine translation used to be easier to identify because it often sounded awkward. Modern AI-generated language can sound polished, natural, and confident. Yet it may still contain serious errors.

That fluency can create false confidence.

A human reviewer working too quickly may assume the output is reliable because it reads well. But accuracy in translation is not only about fluency. It is about meaning, context, purpose, terminology, register, audience, and subject-matter reliability.

Where the source text involves regulated, technical, legal, medical, financial, or safety-critical content, the risks become even higher. In these contexts, the workflow must be transparent and appropriate for the purpose of the translation.

For certification and audit purposes, the question is not only whether the final document reads well. The question is whether the provider can demonstrate that the correct process was followed.

That includes evidence of qualified personnel, defined roles, client instructions, revision or post-editing steps, quality checks, and final approval.

A responsible provider should therefore separate three types of workflows:

This simple distinction can help LSPs avoid misclassification, reduce audit risk, and communicate more clearly with clients.

AI-Supported Workflows Can Still Be Compliant

AI should not be viewed only as a threat to standards. Used properly, AI can support quality, productivity, consistency, and project control. The important point is that AI must be used within a clearly governed process.

For example, an ISO 17100-certified translation service provider may use AI tools to support terminology extraction, style guide checks, formatting review, translation memory maintenance, consistency verification, or pre-delivery quality assurance.

These uses do not necessarily replace the translator or reviser. Instead, they may strengthen the human-led process.

However, the provider must be able to explain and document how AI is used. Auditors may reasonably ask whether AI is used to generate translations, whether clients are informed, whether data confidentiality is protected, and whether human roles remain clearly assigned.

This is where internal governance becomes important.

An LSP should have clear policies explaining when AI tools may be used, what types of content may be processed, what client permissions are required, how confidentiality is protected, and who remains responsible for the final output.

Good AI governance should include internal policies, staff training, risk assessment, client communication, data protection controls, and clear records.

The goal is not to avoid AI entirely. The goal is to use AI responsibly, transparently, and in a way that supports the correct standard.

AI can be valuable when used under professional control. It can help identify inconsistencies, accelerate repetitive checks, support terminology management, and improve workflow efficiency. But it should never create confusion about professional responsibility.

Technology may support quality, but accountability remains human.

Why Client Transparency Matters

Clients should know what type of service they are receiving.

A client ordering human translation and revision may have different expectations from a client ordering machine translation post-editing. Those expectations may affect price, delivery time, confidentiality, quality level, and risk tolerance.

Transparent service descriptions protect everyone involved. They protect the client because the client understands the workflow. They protect the language service provider because the service has been accurately represented. They also protect auditors and certification bodies because the evidence can be assessed against the correct standard.

A clear client-facing description may include:

Human translation and revision in accordance with ISO 17100.

Machine translation post-editing in accordance with ISO 18587.

AI-supported human translation, with qualified human translator and reviser roles maintained.

This kind of clarity helps prevent confusion. It also builds trust. In a market where many clients are uncertain about AI, providers that communicate openly will have a strong advantage.

Client transparency is also important because not all clients have the same risk profile. Some clients may welcome AI-supported workflows for lower-risk content, especially where speed and cost are priorities. Other clients may prohibit AI use due to confidentiality, regulatory, legal, or data protection concerns.

A responsible LSP should not assume that one workflow is suitable for all clients.

The provider should discuss requirements clearly, document client instructions, and ensure that the selected workflow matches the purpose and risk level of the project.

This is especially important where sensitive or confidential information is involved. AI tools may involve data handling considerations, depending on the system used and the provider’s policies. Clients may expect assurance that their content will not be used in ways that compromise confidentiality.

Trust is not built by hiding technology. Trust is built by explaining technology clearly and using it responsibly.

Practical Guidance for Translation Service Providers

Translation service providers should start by mapping their actual workflows. The most important question is simple:

Who creates the first target-language version?

If the first version is created by a qualified human translator, and AI only supports the translator’s work, ISO 17100 may still be appropriate. If the first version is created by AI or machine translation, and the human edits that output, ISO 18587 will usually be the better fit.

Providers should also review their contracts, quotes, project records, internal procedures, and marketing language. It is risky to advertise ISO 17100 compliance for a service that is actually machine translation post-editing.

It is much safer to classify services correctly and apply the right standard to each workflow.

Training is also essential. Translators, revisers, project managers, and post-editors need to understand the difference between human translation, revision, post-editing, and AI-supported linguistic work. Without this understanding, even a well-written procedure may fail in practice.

LSPs should also create practical internal guidance for staff. Project managers should know how to classify jobs. Linguists should know what is expected of them. Sales teams should know how to describe services accurately. Quality managers should know what evidence is needed for audits.

A strong internal process may include:

Workflow classification before project start.

Clear client instructions.

Documented approval for AI or MT use where required.

Defined competence requirements for translators, revisers, and post-editors.

Records showing who performed each step.

Quality assurance checks suitable for the workflow.

Final verification before delivery.

Accurate service descriptions in quotes and contracts.

These controls help ensure that AI is used responsibly and that the service delivered matches the service promised.

For more information about translation standards and certification services, visit Translation Standards:

https://translationstandards.net/

What Auditors and Certification Bodies Should Consider

As AI becomes more common in translation workflows, auditors and certification bodies will need consistent interpretation and practical alignment.

The same workflow should not be treated differently simply because different auditors use different assumptions about AI.

Auditors may need to ask more detailed questions about technology use, workflow structure, client communication, competence records, and quality control.

Useful audit questions may include:

Who produced the first target-language version?

Was machine translation or AI used to generate the translation?

Was the human role translation, revision, review, or post-editing?

Were the professionals involved qualified for the role they performed?

Was the client informed about the workflow?

Are project records clear enough to demonstrate conformity?

Does the LSP distinguish between ISO 17100 and ISO 18587 services?

Are AI-related risks addressed in internal procedures?

These questions are not intended to prevent innovation. They are intended to ensure that innovation does not weaken transparency, quality, or compliance.

The translation industry is evolving, and standards interpretation must evolve with it. However, the core principles remain the same: competence, accountability, clarity, quality, and trust.

Frequently Asked Questions

Does AI translation plus human revision comply with ISO 17100?

Usually, not under ISO 17100 alone if the actual translation is mainly produced by AI and the human role is post-editing. AI may support a workflow, but if it replaces the main human translation role, ISO 18587 may be more appropriate.

Can AI be used in an ISO 17100-certified workflow?

Yes. AI may be used as a support tool for terminology, quality assurance, consistency checks, formatting, or productivity, provided that the required human translator and reviser roles remain clear, active, and demonstrable.

What standard applies to machine translation post-editing?

ISO 18587 is the more suitable standard for full human post-editing of machine translation output. It focuses on the post-editing process and the competences required from post-editors.

What is the difference between revision and post-editing?

Revision normally involves checking a human translation against the source text. Post-editing involves editing and correcting machine translation output. The activities may overlap, but they are not identical.

Should clients be told when AI is used?

Yes. Clients should be clearly informed whether they are receiving human translation, machine translation post-editing, or AI-supported human translation. Transparency supports trust and reduces compliance risk.

Can an LSP offer both ISO 17100 and ISO 18587 services?

Yes. A language service provider may offer both, provided that each service is clearly defined, properly documented, and delivered according to the correct workflow and competence requirements.

Is AI always a risk for translation compliance?

No. AI is not automatically a compliance risk. The risk comes from unclear workflows, poor documentation, weak human control, lack of client transparency, or using the wrong standard for the service delivered.

What is the safest approach for LSPs using AI?

The safest approach is to classify workflows honestly, document the role of AI, maintain qualified human responsibility, inform clients where appropriate, and apply the correct ISO standard to the service being provided.

Conclusion

AI is now a permanent part of the translation industry, but it does not remove the need for professional standards. In fact, it makes standards even more important.

Language service providers must be clear about how translations are produced, who is responsible for each stage, and which standard applies to the workflow.

The most practical conclusion is this: AI can support ISO 17100 workflows, but AI-generated translation followed by human editing is usually better treated under ISO 18587.

ISO 17100 remains the right framework for human translation and human revision. ISO 18587 is the more appropriate framework for machine translation post-editing.

For modern language service providers, the winning approach is not to reject AI. The winning approach is to use AI with transparency, professional judgment, strong documentation, and the correct ISO framework.

This protects clients, supports auditors, and helps the translation industry move forward with both innovation and trust.

Professional standards and artificial intelligence can work together, but only when the workflow is honest, the human role is clear, and the correct standard is applied.

References

Translation Standards
https://translationstandards.net/

ISO 17100 information
https://www.iso.org/standard/89761.html

ISO 18587 information
https://www.iso.org/standard/62970.html