Governance5 min read

AI Governance in 2026: What Changed This Year, and What to Do Before 2027

The EU delayed its toughest AI rules, India chose guidelines over a new law, and US states kept legislating while Washington deregulated. If that sounds like a reason to relax, it isn’t. Here is what actually changed, and what it means for your organisation.

Rows of server racks in a data centre
On this page 7 sections
  1. A year of mixed signals
  2. Europe: the deadline moved, the destination didn’t
  3. India: governance without a new law
  4. United States: deregulation in Washington, rules in the states
  5. The trends underneath the headlines
  6. What to do before 2027
  7. Governance as an advantage

A year of mixed signals

If you only skimmed the headlines in 2026, you could be forgiven for thinking AI regulation went into reverse.

Europe pushed back the deadlines for its most demanding AI rules. India decided not to write a standalone AI law. In the United States, the federal government doubled down on deregulation, while several states pressed ahead with their own laws, and at least one rewrote its law before it ever took effect.

Read more carefully, though, and a clearer picture emerges. The direction of travel hasn’t changed at all. What has changed is the timetable, and the expectation that organisations will govern AI through everyday practice rather than waiting to be told exactly what to do.

This article reflects the position as of 1 November 2026. Regulations and their timelines are changing quickly; check the latest official sources before relying on any date.

Europe: the deadline moved, the destination didn’t

The biggest story of the year was the EU’s Digital Omnibus on AI, adopted in July 2026 as Regulation (EU) 2026/1744. It changed the AI Act’s timetable in important ways:

  • Stand-alone high-risk systems (the Annex III list, which includes uses in employment, credit, education and essential services) now face their obligations from 2 December 2027, not August 2026.
  • High-risk AI built into regulated products (Annex I, such as medical devices and machinery) moves to 2 August 2028.
  • Rules for general-purpose AI models, in force since August 2025, stayed on schedule, and the Commission’s enforcement powers over them apply from August 2026.
  • The AI literacy duty in Article 4 was softened: providers and deployers must now take measures to support AI literacy, rather than ensure a specific level of it.

The stated reason for the delay was practical: the technical standards that high-risk systems are meant to follow weren’t ready. That is a reprieve on timing, not on substance. Anyone deploying AI in hiring, lending, education or customer eligibility decisions in Europe now has roughly a year to get ready, and the work involved (risk management, data governance, documentation, human oversight, logging) takes most organisations longer than they expect.

India: governance without a new law

India took a deliberately different path. In November 2025, MeitY released the India AI Governance Guidelines, which set out seven guiding principles and proposed new coordinating bodies, including an AI Governance Group and an AI Safety Institute, but chose not to create a separate AI law. Instead, AI risks are to be managed through existing legislation, sector regulators and voluntary commitments.

That doesn’t mean there are no hard obligations. The Digital Personal Data Protection Rules were notified the same month, with an 18-month phased rollout. The rules on consent managers apply from November 2026, and the substantive duties, including notice, security safeguards, breach notification and children’s data, apply from 13 May 2027. Most AI systems touch personal data somewhere, so for many Indian organisations the DPDP deadline is effectively their AI governance deadline.

Sector regulators matter too. Financial services firms, for example, should expect AI expectations to arrive through their regulator’s guidance rather than through a single national AI statute.

A delay is a reprieve on timing, not on substance.

United States: deregulation in Washington, rules in the states

At federal level, the direction has been clear since January 2025, when an executive order revoked the previous administration’s AI order. The July 2025 AI Action Plan prioritised innovation, infrastructure and reduced regulation, and was accompanied by executive orders on data centres, AI exports and federal procurement of AI models.

The states have moved the other way. California’s Transparency in Frontier AI Act (SB 53) took effect on 1 January 2026, requiring developers of the largest models to publish safety frameworks and report serious incidents. Colorado, which passed the first broad state AI law in 2024, repeatedly delayed it and in 2026 replaced it with a narrower law focused on automated decision-making, due to take effect in January 2027. New York’s RAISE Act on frontier models is also due in 2027.

For most enterprises, the frontier-model laws won’t apply directly; they target a handful of model developers. But they shape what you can expect from your vendors, and the state laws on automated decisions are much closer to home for anyone using AI in hiring, lending or insurance.

Beyond specific laws, four trends are shaping how AI is governed in practice:

From principles to management systems

Organisations are moving from high-level AI principles to operational management systems. ISO/IEC 42001, the international standard for AI management systems, and the NIST AI Risk Management Framework give a common structure that works across jurisdictions.

Agents raise the stakes

As AI moves from answering questions to taking actions, governance has to cover tool permissions, approval steps and audit trails, not just model outputs. Most existing AI policies were written before agents and need updating.

Procurement is becoming governance

Much enterprise AI arrives inside software you buy. Vendor contracts, model disclosures and the right to audit are becoming central governance tools.

Transparency is becoming the default

Labelling AI-generated content, telling people when they are talking to a machine and explaining automated decisions are spreading across jurisdictions, regardless of how each one regulates the rest.

What to do before 2027

Whatever your jurisdiction, the same practical steps hold up. If we had to pick six things for the next twelve months, they would be these:

  1. Build an AI inventory. List the AI systems you build, buy and embed, including AI features inside existing software. You can’t govern what you can’t see.
  2. Classify by risk. Map each system against the uses regulators care about: employment, credit, education, health, essential services and anything affecting vulnerable people.
  3. Name owners. Every material system needs a business owner and a technical owner who are accountable for it.
  4. Evaluate and log. Test systems before launch against agreed criteria, monitor them after, and keep records that would satisfy an auditor.
  5. Update contracts. Ask vendors about training data, model changes, incident reporting and audit rights, and write the answers into contracts.
  6. Invest in literacy anyway. The EU softened its literacy duty, but people who understand AI remain the strongest control you have.

Map these steps to ISO/IEC 42001 or the NIST AI RMF and you will be in a strong position in most jurisdictions, whatever the next set of amendments brings.

Governance as an advantage

It is tempting to treat every delay as breathing room. The organisations that will benefit most from AI are doing the opposite: using the time to build governance that lets them move faster with confidence, because they know what they are running, who owns it and how it behaves.

Good governance is not the brake on AI adoption. Done well, it is the thing that lets you take your foot off the brake.

Key takeaways

  • EU high-risk obligations now start 2 December 2027 (Annex III) and 2 August 2028 (Annex I).
  • India governs AI through existing law and guidelines; DPDP duties apply from 13 May 2027.
  • US federal policy deregulated while states such as California, Colorado and New York legislated.
  • Build an inventory, classify by risk, name owners, evaluate and log.
  • Use ISO/IEC 42001 or the NIST AI RMF as a common structure across jurisdictions.

Case studies & examples

References & further reading

  1. Digital Omnibus on AI (Regulation (EU) 2026/1744): amended AI ActAI Act Explorer, 2026
  2. Regulation (EU) 2024/1689, the AI ActEUR-Lex, 2024
  3. India AI Governance Guidelines unveiledPress Information Bureau, Government of India, 2025
  4. Government notifies DPDP RulesPress Information Bureau, Government of India, 2025
  5. America’s AI Action PlanThe White House, July 2025
  6. California enacts the Transparency in Frontier AI Act (SB 53)Morrison Foerster, 2025
  7. ISO/IEC 42001:2023 AI management systemsISO, 2023
  8. AI Risk Management Framework and Generative AI ProfileNIST, 2023–2024
  9. OECD AI PrinciplesOECD

Links checked on 1 November 2026. External sources are listed for further reading; Delphilens is not affiliated with them.