On 2 August 2026, key obligations of the EU AI Act for high-risk AI systems take effect. For companies that use AI to translate contracts, technical documentation or medical texts, this is an important deadline. This article explains which translations are affected, what duties arise and how a compliant workflow works in practice.
The EU AI Act is the first comprehensive framework for artificial intelligence. It follows a risk-based approach: the higher the potential harm of an AI system, the stricter the obligations. Spam filters remain largely unregulated. Systems that affect sensitive areas, on the other hand, face clear requirements.
Translation enters the picture wherever machine translation is used for sensitive content. An AI-translated internal note is one thing. An AI-translated contract or package insert is another. The EU AI Act draws a line here that many companies have not consciously drawn before.
Not every translation falls under the strict requirements. The focus is on business-critical and regulated content:
Important for international teams: The EU AI Act also reaches beyond the EU borders. As soon as the AI translation is used inside the EU, the obligations apply, even if the provider is based outside the EU.
The EU AI Act calls for more diligence in how AI translation is handled. Four points are especially relevant for companies:
When AI is used for translation, this has to be disclosed. Customers have a right to know whether a text was produced by a machine.
High-risk AI systems need human control. A qualified person must be able to review results and correct errors before they take effect.
Processes, the tools in use and the review steps should be documented. Who approved which translation needs to remain traceable.
AI systems must be checked for bias and monitored continuously. Companies that train their own translation engines with in-house data should also be able to show where that training data comes from.
Since January 2026, ChatGPT Translate has been available. AI translation is now within reach for everyone: fast, low cost and ready in seconds. The temptation to route everything through such tools is strong.
Yet AI delivers drafts, not responsibility. A mistranslated contract clause, a misleading dosage instruction or a lost meaning in customer communication can have real consequences. The decisive question is therefore not whether AI can translate, but who is responsible for meaning and accuracy in each target language.
The practical path comes down to one simple idea: tier content by risk rather than deciding across the board.
Objective quality benchmarks help make this standard provable. The ISO 17100 standard defines the process and the qualification of translators. The four-eyes principle ensures that a second specialist reviews every translation. And the evaluation standard DIN ISO 5060 makes it possible to assess translation quality, including AI output, against a clear matrix.
The good news: companies that use June and July are prepared in good time. These five steps create clarity:
At tolingo, AI provides the speed and experienced specialist translators carry the responsibility. With certified processes and human sign-off, fast translation becomes reliable communication that holds up in the compliance era.
Discuss your translation needsThe obligations for high-risk AI systems become enforceable from 2 August 2026. Machine translation of sensitive content falls under them.
Mainly legal, medical, official and safety-relevant technical content. Internal texts with low risk are generally less critical.
Yes. What matters is risk-based use: AI for non-critical content, human translation or sign-off for sensitive content, plus transparency and documentation.
Yes, as soon as the AI translation is used inside the EU, the obligations apply, regardless of where the provider is based.
This article offers general guidance on the topic of AI and translation and does not replace legal advice. For a specific legal assessment of your individual case, please consult qualified professional advice.