Decode Civils CAShorts
Current affairs Open app
ExplainerSci-TechGS2GS3

EU AI Act's transparency rules and India's synthetic content labels

By CAShorts DeskPublished Sources: EU AI Act (Official Journal); European Commission; MeitY, IT Rules, 2021 as amended 10 Feb 2026; PIB (MeitY); ThePrint

Why in news: On 5 October 2026 OpenAI said it would add an invisible watermark to ChatGPT and Codex text in the EU to meet Article 50.

EU AI Act's transparency rules and India's synthetic content labels
Illustration (AI-generated)

The European Union's Artificial Intelligence Act, Regulation (EU) 2024/1689, sorts AI systems into four risk tiers, from banned practices to minimal risk. Its Article 50, applicable since 2 August 2026, makes generative AI providers mark output as machine-detectable and deployers disclose deepfakes. India's amended IT Rules require labels on synthetic audio-visual content. OpenAI's EU text watermark brings the rules up.

Syllabus: GS2 · Government Policies; GS3 · IT, Space, Biotech & IPR; GS3 · Cyber Security & Money Laundering

Why this matters for UPSC

GS3 awareness in IT and AI, and the role of social media in internal security: Prelims can ask the AI Act's risk tiers, Article 50 duties or India's synthetically generated information rule; Mains, how to curb deepfakes without stifling innovation.

In short

The EU's Artificial Intelligence Act, Regulation (EU) 2024/1689, regulates AI by risk; its Article 50 makes AI-generated content identifiable, while India relies on intermediary rules under the IT Act.

Four risk tiers of AIActUnacceptable riskHigh riskTransparency riskMinimal risk

India should weigh the EU's provider-level marking duty, extend labelling to text where feasible, and protect speech through proportionate, reviewable takedowns.

Why it matters

A voter who watches a video of a leader saying words never spoken, or a reader who meets an AI-written news summary, needs a way to know what is real. Marking and labelling rules give that signal: a visible tag, hidden data in the file, or a pattern a detector can find. For companies the rules set a cost of doing business: an Indian start-up selling an image tool in Europe must meet the EU's marking duty, and a platform at home must label synthetic videos and act within hours on unlawful ones. The larger question is trust in information when machines can fake faces, voices and text cheaply, weighed against free speech and room to innovate.

Prelims pointers

  • Article 50(1) places the chatbot disclosure duty on providers; it lapses where AI interaction is obvious to a reasonably well-informed person.
  • Article 50(3) requires deployers of emotion recognition or biometric categorisation systems to inform the people exposed to them.
  • For evidently artistic, satirical or fictional works, Article 50(4) limits deepfake disclosure to a manner that does not hamper enjoyment of the work.
  • High-risk AI systems need quality datasets, activity logging, documentation, human oversight, and robustness, cybersecurity and accuracy before they reach the EU market.
  • India's definition excludes routine good-faith editing, document and presentation design, and accessibility or translation uses that do not alter the substance.
  • Rule 3(3) bars an intermediary from enabling removal or suppression of the label or permanent metadata on synthetic content.
  • A reasoned takedown intimation from police must come from an officer not below Deputy Inspector General; a Secretary-level officer reviews intimations monthly.
  • MeitY describes a significant social media intermediary as one with 50 lakh or more registered users in India, a threshold the Centre notifies.

Practise a UPSC-style question

With reference to the risk tiers of the European Union's Artificial Intelligence Act, consider the following statements:

1. AI-enabled video games and spam filters are placed in the high-risk tier and need human oversight.

2. Social scoring is among the AI practices that the Act prohibits.

3. AI used to score exams or to sort CVs for recruitment is classed as high-risk.

Which of the statements given above is/are correct?

  1. (a) 1 and 2 only
  2. (b) 2 and 3 only
  3. (c) 1 and 3 only
  4. (d) 1, 2 and 3
Show the answer

Answer: (b) 2 and 3 only

Answer: (b) 2 and 3 only. Social scoring is banned and exam scoring and CV-sorting are high-risk; games and spam filters are minimal risk.

Statement 1: incorrect. The European Commission places AI-enabled video games and spam filters in the minimal or no risk tier, for which the AI Act sets no rules.

Statement 2: correct. Social scoring is one of the nine practices the Commission lists under the unacceptable-risk tier, which the AI Act bans.

Statement 3: correct. Exam scoring in education and CV-sorting software for recruitment are the Commission's own examples of high-risk use cases.

Eliminate: Statement 2 is the surest, as social scoring is the best-known ban. Out: (c).

Eliminate: Statement 1 is wrong, since spam filters sit in the lowest tier. Out: (a), (d).

Trap: Wrong scope. Statement 1 lifts the commonest, harmless uses of AI into the strictest regulated tier; the Act regulates by the risk of the use, so everyday tools face no new rules.

Remember: Four tiers: banned, high-risk, transparency duties, minimal risk with no rules.

More Explainers

All Explainers

More Sci-Tech current affairs