Sach batao: if your servers, your team and your customers are all in India, why would an EU law have anything to say about your AI?
Because the EU AI Act does not ask where you sit. It asks where the system is sold, and where its output is used. For an Indian software exporter, that can be enough.
This post gives a three-question scope test and the dates as they stand after the EU’s July 2026 amendment. It is a reading of the texts, not legal advice.
๐งญ The Three-Question Scope Test
Article 2(1) of the Act sets out who is covered. According to the AI Act Explorer’s reproduction of the text, three of the groups matter for Indian firms:
- Do you put an AI system or a general-purpose AI model on the EU market? Providers are covered “irrespective of whether those providers are established or located within the Union.” A product you sell to EU customers counts, even from Pune or Bengaluru.
- Do you have a place of establishment or a presence in the EU and use AI there? Deployers established or located in the Union are covered. An Indian company with an EU subsidiary or office is in this group.
- Is the output of your AI system used in the EU? Providers and deployers in a third country are covered where “the output produced by the AI system” is used in the Union. This is the question that catches services firms.
Two exclusions are narrow. Systems built solely for scientific research are out (Article 2(6)), and so is pre-market research and development (Article 2(8)). The same article adds that testing in real-world conditions is not covered by that exclusion.
A “yes” to any of the three does not mean heavy obligations. It means you must work out which role you hold and which rules apply to that role.
๐ The Dates After the July 2026 Amendment
The Act was changed by Regulation (EU) 2026/1744, dated 8 July 2026 and published in the Official Journal on 24 July 2026. It enters into force on the third day after publication, which makes it 27 July 2026.
| What | Date |
|---|---|
| Prohibited practices and Article 4 (Chapters I and II) | Applied from 2 February 2025 |
| General-purpose AI model obligations | Applied from 2 August 2025 |
| Rest of the Act, unless stated otherwise | Applies from 2 August 2026 |
| Some new Article 5 prohibitions added by the amendment | 2 December 2026 |
| Systems already on the market that generate synthetic content: comply with Article 50(2) | 2 December 2026 |
| High-risk systems listed in Annex III | 2 December 2027 |
| High-risk systems under Annex I (product-linked) | 2 August 2028 |
The first three rows come from Article 113 as reproduced on the AI Act Explorer. The last four come from the amending regulation itself on EUR-Lex.
๐๏ธ What Is Still Due on 2 December 2026
The date to watch is narrower than the headlines suggest. A new Article 111(4) says providers of AI systems that generate synthetic content, and that were placed on the market before 2 August 2026, “shall take the necessary steps in order to comply with Article 50(2) by 2 December 2026.”
Article 50(2) is the marking rule. In the text as reproduced, providers of systems that generate synthetic audio, image, video or text must ensure the outputs are “marked in a machine-readable format and detectable as artificially generated or manipulated.”
Two things follow:
- It is a grace period, not a new rule. It applies to systems that were already on the market before 2 August 2026. A system placed on the market after that date is not covered by the extension.
- It applies to providers. If you build a generation feature and sell it, that is you. If you only use someone else’s tool, the marking duty sits with the provider, though other transparency duties for deployers (such as disclosing a deepfake) can still apply.
๐ถ What a Mistake Can Cost
Under Article 99, as reproduced on the Explorer, transparency duties under Article 50 sit in the middle tier: up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher. Prohibited practices go up to EUR 35 million or 7%.
For SMEs and start-ups the rule flips: each fine is capped at the percentage or the amount, “whichever thereof is lower.” That is a ceiling, not a prediction of what any regulator will do.
๐งฎ A Worked Example
This is an illustration, not a real case. Take a 60-person Indian SaaS company that sells a marketing tool to customers in Germany and France. The tool generates product images and ad copy.
- Question 1: yes. It is a provider placing a generation system on the EU market.
- Question 3: also yes, because the output is used in the EU.
- Deadline: if the tool was on the market before 2 August 2026, marking of its synthetic output must be in place by 2 December 2026. If it launched later, the duty already applied from launch.
- Likely not in scope for the high-risk dates. A copy generator is not obviously an Annex III system, so 2 December 2027 is probably not its date, though a lawyer should confirm that.
๐ง One Side Effect Worth Knowing
The amendment also moves the EU Machinery Regulation (2023/1230) from Section A to Section B of Annex I. If you build industrial machines with AI in a safety function, that changes where your product sits in the Act. It deserves its own post, because the dates and obligations differ from the ones above.
โ A Checklist for This Week
- List every AI feature you ship to, or whose output reaches, EU users.
- For each one, write down your role: provider, deployer or both.
- Mark which were on the market before 2 August 2026.
- For synthetic-content features, check how outputs are marked and whether it is machine-readable.
- Ask counsel to confirm whether any feature is high-risk under Annex III.
๐ช Token Truth: an output your system generates in India can become a compliance event the moment someone in the EU uses it.
If an EU customer asked you tomorrow how your AI outputs are marked, could you answer in one sentence?
๐ References
- European Parliament and Council, Regulation (EU) 2026/1744, 8 July 2026, OJ L, 24.7.2026 (citation, entry into force, Article 111(4), Article 113 dates, Annex I changes).
- EU AI Act Explorer: Article 2, Scope (reproduction of the Official Journal text).
- EU AI Act Explorer: Article 50, Transparency obligations.
- EU AI Act Explorer: Article 99, Penalties.
- EU AI Act Explorer: Article 113, Entry into force and application.
- Related on this blog: India’s consent managers and the DPDP Rules.
Checked 21 November 2026. The Article 2, 50, 99 and 113 texts were read from the AI Act Explorer reproduction, not EUR-Lex; the amendment dates were read on EUR-Lex. This is not legal advice.