The EU AI Act rules landing on 2 August 2026, and whether an Indian agency needs to care
On 2 August 2026 a set of EU AI Act obligations becomes enforceable. Most Indian agencies and small businesses are unaffected. A specific minority are affected and mostly do not realise it, so it is worth ten minutes to establish which group you are in.
Two clocks, and people keep reading the wrong one
Coverage has been confusing because the Act's timeline was amended. The European Parliament approved changes in June 2026 that delayed some obligations and left others in place. Law firms tracking it, including Morgan Lewis and Skadden, describe the split clearly.
| Obligation | Date |
|---|---|
| Article 50 transparency duties | Enforceable 2 August 2026 |
| Enforcement and penalty powers over general-purpose AI providers | 2 August 2026 |
| National authorities gain full investigation and sanction powers | 2 August 2026 |
| Standalone high-risk systems (recruitment screening, biometric ID and similar) | Deferred to 2 December 2027 |
| Watermarking obligations for AI content | Moved to December 2026 |
| High-risk AI inside safety-critical products | Deferred to August 2028 |
So the reporting that "the hard parts were delayed" and the reporting that "enforcement begins in August" are both correct, about different provisions.
What Article 50 actually asks for
The transparency rules are, in plain terms, honesty obligations rather than engineering ones:
- Tell people when they are interacting with an AI system, unless it is obvious. A support chatbot that implies a human is answering is the classic problem case.
- Mark synthetic content as machine-generated, in a form that can be detected.
- Label deepfakes and manipulated media depicting real people or events.
- These duties reach deployers as well as providers, and the reference material at artificialintelligenceact.eu notes that open-source systems are not exempt from them.
Notice what this is not. It is not a certification, an audit, or a documentation regime. For most people it is a disclosure line and a labelling habit.
Does it reach you?
The Act applies based on where the system's output is used, not only where the company sits. Work through this honestly:
| Situation | Position |
|---|---|
| Indian business, Indian customers, no EU users | Not in scope. Read this for context and move on |
| You built a chatbot for a client whose users are in the EU | In scope. Disclosure needed |
| Your app is on the Play Store and available in the EU with AI features | Likely in scope. Worth advice |
| You produce AI-generated marketing or media for an EU client | Labelling obligations apply to that content |
| You use AI internally only, staff in India | Not in scope for Article 50 |
| You screen job applicants with AI, EU candidates | High-risk territory, but that timeline moved to December 2027. Do not relax — start documenting now |
If you are in scope, this is a short list
- Inventory where AI touches a user. Chatbots, generated copy, generated images, voice, recommendations. You cannot disclose what you have not listed.
- Add the disclosure. A visible line at the start of a chat: this assistant is automated, ask for a human at any time. Not buried in terms of service.
- Label generated media where it depicts real people, events, or is passed off as a photograph.
- Write down who is responsible. One named person who knows what AI is deployed and where.
- Put it in the client contract. If you build the system and the client operates it, agree in writing who carries the disclosure duty. This is the gap that causes disputes later.
Why this is worth doing even if you are out of scope
Three reasons that have nothing to do with EU enforcement.
First, disclosure is becoming a client procurement question regardless of jurisdiction. Having an answer wins work from firms that do not, which is the same argument as writing an AI use policy before someone asks for one.
Second, regulation tends to converge. India's own framework continues to develop, and teams already labelling AI output will adapt cheaply rather than retrofitting under deadline.
Third, it is simply the honest thing to do. Telling someone they are talking to a machine costs you nothing and preserves the trust that makes the interaction useful at all.
What not to do
- Do not buy an "AI Act compliance platform" because a deadline is near. For Article 50 obligations at small-business scale, this is copy and process, not software.
- Do not assume the December 2027 delay means nothing applies. Enforcement powers and transparency duties start in August 2026.
- Do not rely on a blog post, including this one, as your legal position. If EU users touch your product, get an hour with a lawyer who has read the amended text.
General information, not legal advice. The AI Act is complex, recently amended, and applies differently depending on your role and market. Take advice from a qualified lawyer before relying on any position described here.