Governance

After New Delhi: a one-page map of how the world governs AI

The AI Impact Summit closed with a declaration endorsed by 88 countries and organizations. It sits beside binding laws, voluntary frameworks and targeted rules that a global company now has to read together.

By IAIM Editorial · · Last reviewed · 4 min read

In short: AI is governed through four kinds of instrument: comprehensive binding laws (the EU, South Korea), promotion-first laws and voluntary guidance (Japan, the United Kingdom, Singapore, and the US federal posture), targeted rules (China’s labeling measures, US state laws) and international instruments (the Council of Europe treaty, summit declarations such as New Delhi’s). Underneath, nearly all ask for the same five things, which is why one well-built program can serve many jurisdictions.1,2,3,4,5,6,7

India hosted the AI Impact Summit in New Delhi in February 2026. Its New Delhi Declaration, centered on inclusive benefit from AI rather than on binding rules, was endorsed by 88 countries and international organizations at the close of the summit; later in February India’s Ministry of Electronics and Information Technology reported that three more countries had joined, taking the total to 91.1,8 Declarations do not create obligations for companies, but they show where governments are trying to agree. This is a good moment to step back and map the approaches that do create obligations.

Four models of AI governance

Table: Model, Examples
ModelExamplesBinding on companies?Date to watch
Comprehensive, binding lawEU AI Act; South Korea’s AI Basic Act (in force January 22, 2026)YesEU high-risk rules: August 2026 under the Act, December 2027 under the Commission’s proposed delay2,9
Promotion-first law or voluntary guidanceJapan’s AI Promotion Act; UK AI Opportunities Action Plan; Singapore’s model frameworks; US Executive Order 14365Mostly no; existing law still appliesJapan’s Basic AI Plan; the US federal framework recommendations ordered by EO 143653,4,5,10
Targeted rulesChina’s AI content labeling measures; US state laws on hiring, deepfakes, frontier modelsYes, for the covered useEach rule’s own effective date6
International instrumentsCouncil of Europe Framework Convention; summit declarationsTreaty binds states that ratify it; declarations bind no oneRatifications by the states where you operate7
  • Comprehensive, binding law. The European Union’s AI Act is the reference case; in November 2025 the Commission proposed moving its stricter high-risk obligations from August 2026 to December 2027 as part of a “Digital Omnibus” package, which the Parliament and Council still have to agree.9 South Korea’s AI Basic Act joined it on January 22, 2026, with duties for “high-impact” AI and a domestic-representative requirement for foreign operators.2
  • Promotion-first framework laws and guidance. Japan’s AI Promotion Act, passed on May 28, 2025, sets up an AI Strategy Headquarters and a Basic AI Plan without penalties.3 The United Kingdom has no AI statute; its AI Opportunities Action Plan of January 13, 2025 made 50 recommendations on compute, data and adoption, and existing regulators apply existing law.4 Singapore publishes voluntary model frameworks, most recently one for agentic AI launched on January 22, 2026.5 In the United States, Executive Order 14365 of December 11, 2025 set out to build a national policy framework and to challenge state AI laws the administration considers burdensome; it does not itself regulate companies.10
  • Targeted rules. China regulates through specific measures; its rules on labeling AI-generated content took effect on September 1, 2025, with a mandatory national standard for visible and embedded labels.6
  • International instruments. The Council of Europe’s Framework Convention on AI, the first legally binding international AI treaty, opened for signature on September 5, 2024, with the EU, the United Kingdom and the United States among the first signatories.7

Our earlier briefings cover several of these in depth: Korea’s Act, Singapore’s agentic framework, what EO 14365 does and does not do, and the EU rules for general-purpose models.

The convergence underneath

For a manager, the differences matter less than what the models share. Almost every one asks for the same things in different words: notice to the people affected, an assessment of impact before consequential use, documentation, human oversight of decisions that matter, and a way to report and fix incidents. A program built to that common standard turns each new jurisdiction into a mapping exercise rather than a new project. Frameworks such as the NIST AI RMF and ISO/IEC 42001 are built around exactly those elements, which is why they travel well.

A practical habit

Keep a one-page jurisdiction register for your AI systems: where the affected people are, which instrument applies there (binding law, guidance or none), the obligation that matters most, and the next date to watch. Review it each quarter, and add a line whenever a product launches in a new market. The register is one of the working documents built in Module 12 of the AIMP program.

Most of the work in global AI compliance is noticing change early enough to plan for it. The New Delhi Declaration changes no one’s obligations this year; the register is how you will notice the instruments that do.

Sources

  1. All India Radio News, "India AI Impact Summit 2026 adopts New Delhi Declaration" (February 2026) (opens in a new tab) newsonair.gov.in
  2. Cooley, "South Korea's AI Basic Act: Overview and Key Takeaways" (January 27, 2026) (opens in a new tab) cooley.com
  3. White & Case, "Japan's first AI legislation becomes law" (2025) (opens in a new tab) whitecase.com
  4. Mishcon de Reya, "UK Government launches AI Opportunities Action Plan" (January 2025) (opens in a new tab) mishcon.com
  5. CMS, "Singapore issues global-first agentic AI model governance framework at Davos" (January 29, 2026) (opens in a new tab) cms.law
  6. Loeb & Loeb, "China's AI Labeling Measures and Mandatory National Standards Take Effect September 1" (March 2025) (opens in a new tab) loeb.com
  7. Council of Europe, "Council of Europe opens first ever global treaty on AI for signature" (September 5, 2024) (opens in a new tab) coe.int
  8. Ministry of Electronics and Information Technology (Government of India), Digital India press release, "More countries join the New Delhi Declaration on AI Impact" (February 2026) (opens in a new tab) digitalindia.gov.in
  9. Euronews, "European Commission delays full implementation of AI Act to 2027" (November 19, 2025) (opens in a new tab) euronews.com
  10. Federal Register, Executive Order 14365, "Ensuring a National Policy Framework for Artificial Intelligence" (December 11, 2025), 90 FR 58499 (opens in a new tab) govinfo.gov

IAIM articles are educational and are not legal advice.

Newsletter

Get the IAIM Briefing

A monthly summary of what changed in AI regulation, governance and practice, with the sources for every item. Free. Unsubscribe in one click. See a sample issue.

We send a confirmation email first. See the privacy policy.

Earn the AIMP certificationTwelve modules, a capstone plan and a timed exam. $695 to enroll ($595 for IAIM Members), with a verifiable certificate when you pass.

About the AIMP AI management certification