Definition
The same chatbot may be legal to launch in one country, need a risk assessment in another and be restricted in a third. AI law is developing fast around the world, and anyone building AI needs to know the main approaches.
AI laws and policies are the rules, standards and government strategies that govern how AI is developed and used. Approaches include comprehensive risk-based laws such as the European Union's AI Act, sector-specific rules for finance, health or elections, data-protection laws such as India's DPDP Act that apply to AI using personal data, technical standards such as ISO/IEC 42001, voluntary principles and guidelines, and national strategies that fund research and computing. Most frameworks scale obligations to the level of risk an AI system poses to people's safety and rights.
In simple words
Traffic rules are stricter for trucks than for bicycles, because trucks can cause more harm. Risk-based AI laws work the same way: a game recommender faces few rules, a hiring or medical AI faces many, and some uses are banned.
Overview
- AI laws and policies set rules for how AI may be built and used.
- The EU AI Act, in force since 1 August 2024, is the first broad AI law and uses a risk-based approach.
- India regulates AI mainly through existing laws such as the IT Act and the DPDP Act 2023, plus guidelines.
- Other approaches include US executive actions and state laws, China's rules on generative AI and the UK's principles-based approach.
- Existing laws on discrimination, consumer protection, copyright and liability also apply to AI.
- International frameworks include the OECD AI Principles and UNESCO's 2021 Recommendation.
Why it matters
Laws decide what you may build, what you must test and document, and what happens when things go wrong. Knowing them protects users, avoids penalties and helps you design responsibly from the start.
Key terms
- Risk-based regulation
- Rules whose strictness depends on how much harm an AI use could cause.
- High-risk system
- Under the EU AI Act, AI used in areas such as hiring, credit, education, essential services or law enforcement, with strict duties.
- Conformity assessment
- Checking and documenting that a system meets legal requirements before it is sold or used.
- General-purpose AI model
- A model, such as a large language model, usable for many tasks, with its own transparency duties under the EU AI Act.
- Regulatory sandbox
- A supervised setting where new AI products are tested with regulators before full launch.
- Deepfake labelling
- Requirements to disclose that content was generated or altered by AI.
Types
AI regulation comes in several types. These are the main ones you will meet.
1. Risk-based AI laws
Laws that set stricter rules for AI uses with higher risk to people.
- How it works: The EU AI Act bans unacceptable-risk uses, imposes strict duties on high-risk systems, requires transparency for limited-risk systems and leaves minimal-risk systems largely free.
- Example: Social scoring by governments is banned; AI used in hiring is high-risk and needs documentation and human oversight.
- When to use it: Learn them if your product is used in the EU, since they apply regardless of where the company is based.
2. Data protection laws
Laws on collecting and processing personal data, which cover most AI systems.
- How it works: They require a lawful basis or consent, purpose limits, security, rights to access and erase data, and rules on automated decisions.
- Example: India's DPDP Act 2023 allows penalties of up to ₹250 crore for some breaches.
- When to use it: Follow them for every dataset that contains personal information.
3. Sector-specific rules
Rules from regulators for particular areas such as health, finance, transport and elections.
- How it works: Regulators adapt existing approval, testing and reporting rules to AI products in their sector.
- Example: AI-based medical devices needing approval from medical regulators before hospitals can use them.
- When to use it: Check them for any product in a regulated industry.
4. Copyright and content rules
Rules about training on protected works and about AI-generated content.
- How it works: Courts and lawmakers are deciding when training is allowed, whether AI outputs can be copyrighted and when deepfakes must be labelled.
- Example: India's 2025 proposals to require labels on AI-generated images and videos.
- When to use it: Follow them when using others' content for training or publishing AI-generated media.
5. Voluntary standards and guidelines
Non-binding principles, technical standards and codes of practice.
- How it works: Organisations adopt frameworks such as the NIST AI Risk Management Framework (2023) and ISO/IEC 42001 (2023) for AI management systems.
- Example: A start-up following NITI Aayog's Responsible AI principles to win government contracts.
- When to use it: Use them to build good practice before laws require it and to show customers you are trustworthy.
Explained in detail
Classify the decision
Imagine a school buys one product with two features. One suggests spelling corrections in practice essays. The other ranks applicants for admission. Start a separate record for each feature. Write who makes the decision, how the output affects a learner and whether a person can review it. Then look up the intended use, relevant exceptions and application date in the official text. A vendor's description of its product as educational misses this distinction.
The EU AI Act
The EU AI Act entered into force in August 2024 and applies in stages. Classification depends on the intended use. A timetable helper and a system used to decide admission to a school have different implications. Certain practices have been prohibited since February 2025, including emotion inference in education or workplaces apart from medical or safety purposes. Under the July 2026 amendment, high-risk rules for listed uses such as admissions or recruitment apply from 2 December 2027. The corresponding rules for specified regulated products apply from 2 August 2028. Transparency duties, including telling people when they interact directly with AI unless this is obvious, have their own dates and exceptions. Check the consolidated text for the particular use.
Other approaches around the world
| Country or body | Approach |
|---|---|
| India | DPDP Act 2023 for personal data; IT Act and IT Rules for online platforms, including deepfake and misinformation concerns; NITI Aayog Responsible AI principles; the IndiaAI Mission, approved in March 2024 with an outlay of about ₹10,372 crore, funds computing, data sets, skills and startups |
| China | Specific rules for recommendation algorithms (2022), deep synthesis (2023) and generative AI services (2023), including content controls and labelling |
| United States | Mostly sector regulators and state laws; federal policy has shifted between administrations |
| International | OECD and UNESCO principles; Council of Europe Framework Convention on AI (2024); international AI safety summits; ISO/IEC 42001 management standard |
Laws are only part of governance. Courts apply existing laws on discrimination, consumer protection, defamation and copyright to AI cases, and regulators such as the Reserve Bank of India issue guidance for AI in their sectors. Because rules change quickly, check current official sources before launching a product, and design systems that can produce the documentation, testing evidence and human oversight that almost every framework expects. Building to a high standard from the start is easier than retrofitting compliance later.
How it works
- Identify where the system will be used and which laws apply there.
- Classify the use by risk: is it banned, high-risk, transparency-only or minimal?
- Check data-protection duties for any personal data.
- Prepare documentation, testing evidence, human oversight and user notices.
- Monitor changes in law and update the system and documents.
Try it yourself
EU AI Act risk tier
Which tier does each use most likely fall into?
Government social scoring of citizens
AI that ranks job applicants
A customer-service chatbot
Emotion recognition to monitor pupils in school
A spam filter
Match the law or policy to its focus
What does each mainly cover?
- DPDP Act 2023
- EU AI Act
- IndiaAI Mission
- ISO/IEC 42001
Common mistakes
- Assuming AI is unregulated because there is no single "AI law" in a country.
- Ignoring existing laws on discrimination, consumer protection and data.
- Relying on old information when rules change quickly.
Check yourself
Summary
- AI is governed by risk-based laws, sector rules, data-protection laws, standards and principles.
- The EU AI Act bans some practices and sets strict duties for high-risk systems; India relies on the DPDP Act, IT rules, principles and the IndiaAI Mission.
- Check current official sources and build documentation, testing and oversight in from the start.
Finished reading?
Sources: EUR-Lex, consolidated AI Act with 2026 amendments · European Commission: the EU AI Act · Digital Personal Data Protection Act, 2023 (MeitY) · IndiaAI Mission (MeitY)