Since 2 August 2026, Spain’s AI Supervisory Agency (AESIA) has been able to carry out inspections. Not issuing threats, not sending warnings: inspecting. And if it finds that your company uses AI without an inventory of systems, without training records or without any kind of documented human oversight, it can open proceedings.
This isn’t science fiction. The AI Act has been in force since August 2024. What changed this August is that the training obligation under Article 4 became fully enforceable, with fines of up to 15 million euros or 3% of global annual turnover, whichever is higher.
And here’s the fact that surprises a lot of people: if your company uses DeepL to translate a quote, ChatGPT to draft a technical data sheet or any MT (machine translation) tool to handle internal documentation, your company is already an AI deployer under the regulation. With everything that entails.
Quick summary: The AI Act (EU Regulation 2024/1689) became fully applicable on 2 August 2026. From that date, Spain’s AI Supervisory Agency (AESIA) can inspect and fine companies that use AI without the required documentation. The obligations include an inventory of AI systems, training records for each employee and demonstrable human oversight. Companies that use machine translation tools (DeepL, ChatGPT, MT built into their ERP) are deployers under the regulation. At Overseas Translations, we have been managing translation projects for more than 13 years with documented processes and ISO 17100:2015 certification, which makes us a provider that can demonstrate human oversight on every delivery.
What the AI Act is and what changed on 2 August
The European Artificial Intelligence Regulation (EU Regulation 2024/1689) is the world’s first comprehensive legal framework regulating the use of AI according to levels of risk. It entered into force on 1 August 2024, but its application deadlines are staggered.
The timeline is as follows:
- February 2025: the bans on unacceptable AI practices and the training obligation under Article 4 came into effect.
- August 2025: the governance rules and the obligations for general-purpose AI (GPAI) models, the large language models such as GPT, began to apply.
- 2 August 2026: full application for high-risk AI systems. AESIA gained effective powers to inspect and impose penalties.
Article 4, the one that most affects the majority of mid-sized companies right now, has been enforceable since February 2025. In short, it says that those who deploy AI must ensure that the people using it have sufficient competence, knowledge and training to use it in their context. And that they can prove it.
2 August was not when the obligation “started”: it was when AESIA’s ability to check whether you have complied started.
Is your company an AI deployer? Probably yes
The AI Act distinguishes between providers, who develop the AI system, and operators or deployers, who use it in their professional activity. If your company uses any AI tool to generate, review or manage content in a business context, you are a deployer.
This includes:
- DeepL Pro to translate emails with foreign customers or product sheets.
- ChatGPT or any GPT to draft, summarise or review documents.
- Copilot built into Microsoft 365.
- MT systems built into your ERP, PIM or export platform.
- Any customer service software with automated replies in another language.
According to data from Wolters Kluwer and BBVA Research (2026), 76% of Spanish SMEs use AI every week. Only 8% have an AI solution with documented governance. The remaining 68% use shadow AI, with no inventory, no internal policy and no records of any kind.
That is exactly what AESIA will find if it opens an inspection.
What you need to have documented since 2 August
There is no official format, but there is a clear criterion: reasonable traceability. If you are inspected, you need to be able to show three things:
- An inventory of AI systems: which tools are used in the company, what for and by whom.
- Training records: who has been trained, when and on what.
- Documented human oversight: who reviews the AI output in processes where it has real effects.
The third point has the biggest implications for anyone handling technical, legal or commercial documentation in other languages.
What this means if you use AI to translate
This is where many companies have a real gap.
Imagine your company exports machinery to Germany. You translate the technical manual with DeepL, someone in the export team who speaks a bit of German checks it, and you send it off. If there’s an accident and proceedings are opened, can you prove that the manual was reviewed by a qualified professional? Do you have a record of who reviewed it, when and against what criteria?
It’s the same with tax returns. You don’t hire an accountant because AI can’t do the sums. You hire one because the person who signs is liable if there are errors, and you want someone with sound judgement to have checked the work.
Exactly the same applies to the documents your company sends out in other languages. AI translates well in many cases. The problem isn’t that it makes mistakes: it’s that if it gets something wrong in a contract, a technical data sheet with safety information or a quote with commercial terms, you are liable under product liability law, and the new European Directive, due to be transposed in December 2026, explicitly covers AI software. The AI Act adds another layer: since 2 August, you must be able to show that the people using these tools in your company understand their limits and that there is a review process for the documents that really matter.
How we handle it at Overseas Translations
For more than 13 years, we have been managing translation projects for companies exporting to demanding markets. Since 2017, we have been certified to ISO 9001:2015 and ISO 17100:2015, the specific standard for professional translation projects. This means every delivery follows a documented process: who translated it, who reviewed it and what criteria were applied.
In practice, what we offer our clients is the layer of human oversight the AI Act requires for the documents that really matter: translations using MTPE (professional post-editing) carried out by native translators specialising in your sector, with full traceability and ISO certification behind them.
We’re not against the MT tools your company already uses. In fact, we can work with them. What we offer is the assurance that whatever goes out of your company (to your customers, your distributors, the authorities in the destination country) has the appropriate level of review, and that you can prove it if anyone asks.
Frequently asked questions about the AI Act
Does the AI Act affect SMEs or only large companies?
It affects any company that operates in the EU and uses AI systems in its professional activity, regardless of size. SMEs have the same training and documentation duties, although the most demanding obligations mainly apply to high-risk systems.
When can AESIA start imposing fines?
AESIA has had inspectors in operation since 2025. Its full powers to inspect and impose penalties, including in relation to Article 4 training, have been in effect since 2 August 2026.
Does using DeepL or ChatGPT in my company make me an AI deployer?
Yes. If you use these tools in a professional context within the EU, you are a deployer or operator under the AI Act, with the corresponding obligations: an inventory of systems, documented training and human oversight in processes where the AI’s output has real-world effects.
Does post-editing translations meet the human oversight requirement?
Professional post-editing (MTPE) carried out by a native translator specialising in the sector is the most thorough way to demonstrate human oversight in AI-assisted translation. Even more so if the provider is certified to ISO 17100:2015.
What documents do I need to have ready now that the AI Act is fully applicable?
At a minimum: an inventory of the AI systems in use and training records showing who has been trained, when and on what. For critical documents in other languages, add a record of human oversight for each delivery.
Sources: Regulation (EU) 2024/1689 — official text · AESIA · Data on AI adoption among SMEs: Wolters Kluwer + BBVA Research (2026)
