Europe Wrote the Rulebook, Then Blinked. India Still Undecided

The European Union wrote the world's first comprehensive AI governance framework but then postponed its toughest part, meanwhile India hasn't written one at all. On 27 July 2026, with barely a week to spare, the EU rewrote its one deadline. Hiring and credit-scoring tools now have until December 2027, and AI built into regulated products like medical devices until August 2028 [1].

India took the opposite route. When MeitY released the India AI Governance Guidelines in November 2025, it chose guidelines over a statute. The drafting committee concluded that a separate AI law was “not required at this stage,” because current laws and timely enforcement could address most risks [2].

Imagine two founders are building the same hiring tool, one in Berlin, one in Bengaluru, with one racing towards a rulebook, though still unfinished, and the other has principles and a few general laws. Which is safer? No one knows yet.

What the EU built?

The Act sorts AI by risk and imposes penalties. It carries fines of up to €35 million or 7% of global turnover for banned practices [3]. Its supporters see legal certainty and firm red lines. Even the former Italian Prime Minister Mario Draghi, who later urged a pause [4], affirmed that the early bans on unacceptable-risk systems were implemented smoothly [5].

However, the Act has received a widespread backlash too. A 2022 survey by appliedAI and partner groups found a third to half of polled startups expecting to count as high-risk, against the Commission's assumption of 5 to 15 percent, and nearly three in four investors expecting weaker competitiveness [6]. In 2025, over 40 CEOs, including Airbus' and Philips', asked for a two-year delay [7], while 50-plus civil-society groups warned that such a delay would weaken accountability [8]. Reports also cited missing compliance tools and standards [9].

What others are doing?

The comprehensive legislation enacted by South Korea took effect on 22nd January 2026. Fines under it top out near KRW 30 million (about 21000 USD), after a grace period of at least a year [10][11]. Japan's 2025 law sets no penalties [12]. The UK still has no AI bill, offering sandboxes instead [13]. China has no single AI law but targeted rules which provide for mandatory AI-content labelling since September 2025 [14]. In the USA, a December 2025 executive order targeted state AI laws but does not itself override them [15], and states passed 109 AI laws in the first half of 2026 [16].

Most of these regimes are under two years old. Nobody has earned a verdict yet.

India's third way

On 5 November 2025, India's answer to what AI rules it needs was guidelines, not a statute: seven principles called sutras, adapted from the Reserve Bank's AI framework, and a bet that existing laws could cover most risks [2][17]. Firms would commit voluntarily first, with mandatory rules considered if that proved inadequate over nine to twelve months [18].

Where harm was sharp, India moved fast. Since 20 February 2026, platforms must label AI-generated audio and video, and act on government or court takedown orders within three hours, down from 36 [19]. Data protection duties phase in from May 2027 [20]. A copyright proposal would license AI training on lawfully accessed works in return for royalties [21]. The state is also building up computing capacity, with 38,000 GPUs offered at Rs 65 an hour [22]. Demand in India is real as India has 100 million weekly ChatGPT users, according to Sam Altman [23].

The catch

Critics say voluntary rules run on goodwill. The Internet Freedom Foundation argues that oversight bodies sit inside the IT ministry without statutory independence and that binding safeguards are missing [24][25]. Others fear the deepfake rules' vague definitions could invite over-removal of lawful content [26]. Nasscom, though, praised an innovation framework that respects market realities [27]. Both sides have a point.

The turn

However, the narrative is changing. As recently as December 2025, the plan was to lean on existing law. In July, IT Secretary S. Krishnan said the government would begin stakeholder talks and drafting on AI regulation [28]. On 8 October, Minister Ashwini Vaishnaw promised a regulation-focused consultation paper within a month, while noting that law alone can't regulate technology [18]. What any statute would look like remains unknown.

What should we push for?

We can initiate a few steps for better results. First, binding a few high-stakes areas, such as credit, hiring, welfare and policing; Korea's high-impact list already covers loan screening [11]. Second, building the referees before the rulebook, since Europe's delay is an example of rules stalling without tools and standards [9]. Third, settling data and copyright questions early; courts are already filling gaps, as the Delhi High Court did in July by denying ANI interim relief against OpenAI [28]. Fourth, use sandboxes and align with international standards so exporters don't comply twice.

The bottom line

Europe's lesson isn't that rules fail but that rules need certain machinery. India's lesson so far is that flexibility is easy to praise when nothing binds you. With a consultation paper due within weeks, we'll soon learn whether India's approach was a strategy or postponement.

Europe Wrote the Rulebook, Then Blinked. India Still Undecided

Europe Wrote the Rulebook, Then Blinked. India Still Undecided

What the EU built?

What others are doing?

India's third way

The catch

The turn

What should we push for?

The bottom line

Sources

Let's make your AI assured.