Law, Politics and Policy
Last reviewed: November 2026
AI has become a question of law, power and geopolitics as much as technology. Governments are racing to regulate it without stifling it, courts are applying old laws to new problems, and the world’s major powers are competing to set the rules and control the chips, data and energy that AI depends on.
This page is a plain-English guide to where things stand. Each section opens with a short answer, followed by the detail and links to sources in the footnotes (“References”); some sources are behind paywalls. Because law and policy change often, cases and rules carry their dates. This page is not legal advice. For the latest developments, see the AI Forum Weekly Briefing.
Key points
- There is no single global approach: the EU legislates, the UK relies on existing regulators, the US favours light federal rules, and China combines state direction with control of information.
- Responsibility for AI’s mistakes still rests with the people and organisations that build, deploy and use it, including professionals who rely on its output.
- Copyright is the most contested legal question: whether training AI on protected works is lawful, and whether AI-generated work can be protected at all.
- Control of chips, energy and data has made AI a central arena of competition between the US and China.
How is AI regulated around the world?
In short: four broad models have emerged, reflecting different political values: comprehensive rules in the EU, a regulator-led approach in the UK, federal deregulation in the US, and state-directed control in China.
- European Union, rules-based: the EU AI Act is the world’s first comprehensive AI law. It sorts AI uses by risk, bans some practices outright and imposes strict obligations on “high-risk” uses. It applies to any AI system offered in the EU, wherever the provider is based, so its influence reaches well beyond Europe 1.
- United Kingdom, regulator-led: rather than one AI law, existing regulators such as the ICO, FCA, CMA and Ofcom apply five principles (safety, transparency, fairness, accountability and redress) within their own sectors 2. The government has said it will legislate in some form, but the May 2026 King’s Speech contained no standalone AI bill 3. Private Members’ Bills calling for a dedicated AI regulator have not become law 4.
- United States, deregulatory: federal policy prioritises speed and global leadership. America’s AI Action Plan (July 2025) set out that aim 5, and a December 2025 executive order seeks to establish a single national framework and challenge state AI laws that conflict with it 6. Voluntary standards such as NIST’s AI Risk Management Framework fill much of the gap 7.
- China, state-directed: China treats AI as strategic national infrastructure, with a long-standing goal of world leadership by 2030 8. It has binding rules for recommendation algorithms, deepfakes and generative AI, which combine safety requirements with control of content 9.
What does the EU AI Act require, and when?
In short: the Act is phased in over several years. The bans and the rules for general-purpose AI models already apply. In 2026 the EU amended the Act to give organisations longer to meet the high-risk requirements.
- February 2025: bans on “unacceptable risk” practices, such as social scoring and manipulative techniques, took effect 10.
- August 2025: obligations for providers of general-purpose AI models, such as documentation and copyright policies, began to apply.
- August 2026: transparency duties apply, for example telling people when they are interacting with AI and labelling AI-generated content.
- December 2027 and August 2028: under the “Digital Omnibus” amendments, which came into force in July 2026, the main high-risk obligations now apply from 2 December 2027 for standalone high-risk systems (for example in recruitment, credit or education) and from 2 August 2028 for AI built into regulated products 11 12.
Who is liable when AI goes wrong?
In short: mostly the same people who would be liable without AI. Courts are applying existing law on negligence, contracts and product safety, and responsibility is shared along a chain from developer to deployer to user.
- A chain of responsibility: when an AI system causes harm, the developer, the organisation that deployed it and the person who relied on it may each bear some responsibility, depending on what they knew and could control 13.
- Product liability: the EU’s revised Product Liability Directive (2024) treats software, including AI, as a product, making it easier for people harmed by defective AI to claim compensation 14.
- Insurance: insurers are developing cover for AI-related risks and asking clients how they govern AI, which in practice shapes how businesses deploy it 15.
- Boards and directors: oversight of AI risk is increasingly seen as part of directors’ duties, with implications for directors’ and officers’ liability 16.
AI and intellectual property
In short: whether AI can lawfully be trained on copyright works is still being fought out in courts and parliaments. Meanwhile, the law is clear that AI itself cannot be an author or an inventor, and in the UK it has become easier to patent AI-based inventions.
- Training data (UK): in March 2026 the government reported that a broad copyright exception for AI training, with an opt-out for rights holders, was no longer its preferred approach. It set no replacement and said it would gather more evidence 17.
- Getty Images v Stability AI (High Court, November 2025): in the first major UK ruling on generative AI and copyright, the court held that the trained model was not an “infringing copy” of Getty’s images because it did not store them. Getty succeeded only on a very limited trade mark point. The central question of training in the UK was not decided, because the training took place abroad 18.
- Training data (US): the question turns on “fair use”, and the US Copyright Office has said that some training uses will qualify and others will not, depending on the facts 19.
- AI-generated works: in the US, a federal appeals court confirmed in 2025 that a work created by AI without human creative input cannot be protected by copyright 20.
- AI as inventor: in Thaler v Comptroller (2023), the UK Supreme Court held that only a person can be named as an inventor on a patent 21.
- Patenting AI: in Emotional Perception (February 2026), the UK Supreme Court overturned decades of case law, abandoning the “Aerotel” test and bringing the UK’s approach to patents for AI and software into line with the European Patent Office 22.
Data protection and automated decisions
In short: existing data protection law already applies to AI. Organisations need a lawful basis for using personal data to train and run AI, must be transparent about it, and must give people safeguards when significant decisions about them are automated.
- Training and deploying models: European data protection authorities have set out how GDPR principles such as lawfulness, purpose limitation and data minimisation apply to AI models 23.
- Automated decisions: where AI makes decisions with legal or similarly significant effects, such as on credit or employment, people have rights to information, to human involvement and to challenge the outcome 24.
- Impact assessments: high-risk AI processing generally requires a data protection impact assessment before it begins.
AI in legal practice
In short: AI is now widely used in law firms, but courts and regulators have made clear that lawyers remain fully responsible for everything they put before a court, and that citing AI-invented cases can lead to serious consequences.
- Fake citations: in Ayinde v Haringey (June 2025), the High Court warned that lawyers who cite fictitious, AI-generated authorities risk contempt proceedings and referral to their regulator, and that firm leaders must put measures in place to prevent it 25.
- Judicial use: the judiciary has issued guidance on how judges may and may not use AI, stressing confidentiality and the need to check any output 26.
- Regulators’ expectations: the Solicitors Regulation Authority expects firms using AI to protect client confidentiality, supervise outputs and remain accountable for the service they provide 27.
For how AI is reshaping the business of law and other professional services, see the AI Futures Forum’s research reports for professional service firms.
Deepfakes, elections and content provenance
In short: AI can now produce convincing fake images, audio and video cheaply. Governments are responding with labelling rules and targeted bans, and industry with technical standards for showing where content came from.
- Different legal approaches: China requires AI-generated content to be labelled; Western jurisdictions tend to target specific harms, such as non-consensual intimate images and election manipulation, while protecting free expression 28. The EU AI Act’s transparency duties add labelling requirements from August 2026.
- Elections: electoral authorities have issued guidance to voters and campaigners on AI-generated material 29.
- Provenance and watermarking: the C2PA standard attaches a tamper-evident record of how a piece of content was created and edited 30, while invisible watermarks such as Google’s SynthID mark AI-generated content at the point of creation 31.
The geopolitics of AI
In short: AI has become central to competition between the US and China. The US uses its control of advanced chips as leverage, China is pursuing self-sufficiency, and other countries are deciding whose technology and rules to adopt.
- The chip chokepoint: the US restricts exports of advanced AI chips and chip-making equipment to China 32. The controls have slowed Chinese AI developers but also spurred investment in domestic alternatives and more efficient models 33. In 2025 the US agreed to let some chip sales to China resume in return for a share of the revenue, a new, transactional turn 34.
- Open-source models: freely available models spread innovation and check the power of the largest firms, but can also be adapted for misuse. Chinese developers have become leading producers of open models, a source of global influence 35.
- AI nationalism: governments increasingly treat AI as a strategic industry, using subsidies, data-centre investment and control of key assets 36.
- The rest of the world: how countries in the Global South choose between US and Chinese technology and governance models may shape the long-term balance of power 37.
Energy and resources as strategic assets
In short: AI’s appetite for electricity has turned power supply into a strategic issue. Countries and companies are competing for energy, grid connections and sites for data centres.
- Demand: the International Energy Agency projects that data-centre electricity use will roughly double by 2030 to around 945 TWh, slightly more than Japan uses today, with AI the main driver 38.
- National strategies: the US treats energy and data-centre build-out as a security priority. China’s “East Data, West Computing” programme places data centres in western regions with plentiful renewable power 39. Large technology firms are investing in nuclear power, including small modular reactors.
- Environmental reporting: Europe is pressing for disclosure of data centres’ energy and water use, making sustainability part of the competition 40.
Democracy, power and public trust
In short: the cost of building frontier AI concentrates power in a handful of companies and states, AI can be used to inform or to manipulate citizens, and public attitudes to AI differ sharply between countries.
- Concentration of power: frontier AI requires vast capital, data and computing power, which risks concentrating economic and political influence in very few hands 41.
- Government use of AI: opaque AI systems in public administration raise questions of transparency, accountability and how citizens can challenge decisions 42.
- Public attitudes: surveys find people in China and many emerging economies far more optimistic about AI’s benefits than people in the US, UK and much of Europe, where concerns about jobs, misinformation and privacy act as a brake on adoption 43.
International cooperation
In short: despite the rivalry, countries continue to meet on AI, though the focus has shifted from safety towards opportunity and access.
- The summit series: the 2023 Bletchley Declaration was the first agreement among leading AI nations, including the US and China, on the risks of frontier AI 44. Later summits in Seoul (2024), Paris (2025) and New Delhi (2026) moved the emphasis towards innovation and access; the New Delhi Declaration was endorsed by around 90 countries and organisations 45.
- Testing institutes: the UK’s AI Security Institute (originally the AI Safety Institute) and similar bodies elsewhere test advanced models for dangerous capabilities 46.
- Shared principles: the OECD AI Principles, adopted by dozens of countries, remain the most widely shared baseline for trustworthy AI 47.
Key terms
- EU AI Act: the EU’s comprehensive, risk-based law regulating AI.
- Export controls: government restrictions on selling sensitive technology, such as advanced chips, to other countries.
- Fair use: the US legal doctrine allowing some use of copyright works without permission.
- General-purpose AI model: a model that can be used for many different tasks, such as a large language model.
- High-risk AI: under the EU AI Act, uses of AI that could significantly affect safety or fundamental rights, such as in recruitment or credit.
- Provenance: a verifiable record of where a piece of content came from and how it was changed.
- Text and data mining: automated analysis of large volumes of text or data, including copying works to train AI.
For more definitions, see the full AI Glossary.
Follow developments
Laws, cases and policies in this area change frequently. For the latest, see the AI Forum Weekly Briefing. Related primers: Philosophy, Ethics and AI Alignment, Technology and Science, and Economy and Society.
References
- EU AI Act: first regulation on artificial intelligence – European Parliament
- AI regulation: a pro-innovation approach – GOV.UK
- King’s Speech signals diffuse UK digital policy agenda, but no AI bill – IAPP
- AI regulation in the UK: debate on the need for cross-sector legislation – House of Lords Library
- America’s AI Action Plan – The White House
- Executive Order 14365: Ensuring a National Policy Framework for Artificial Intelligence – The American Presidency Project
- AI Risk Management Framework – NIST
- China’s New Generation Artificial Intelligence Development Plan (2017) – DigiChina, Stanford
- China’s AI Policy in the DeepSeek Era – Carnegie Endowment
- AI Act – European Commission
- EU AI Act Omnibus Agreement: Postponed High-Risk Deadlines and Other Key Changes – Gibson Dunn
- The Digital Omnibus on AI enters into force today – Lewis Silkin
- AI liability: who is accountable when artificial intelligence malfunctions? – Taylor Wessing
- Directive (EU) 2024/2853 on liability for defective products – EUR-Lex
- Trends in AI Insurance Coverage and Claims Handling – Risk Management Magazine
- AI in the Boardroom: Navigating New Risks for Directors and Officers – Marsh
- Opted out: UK government moves away from preferred position on AI and copyright – Osborne Clarke
- Stability AI defeats Getty Images copyright claims in first-of-its-kind dispute – Bird & Bird
- Copyright Office Issues Key Guidance on Fair Use in Generative AI Training – Wiley
- Thaler v. Perlmutter (D.C. Cir. 2025) – Justia
- Thaler v Comptroller [2023] UKSC 49: AI cannot be an inventor – RPC
- Emotional Perception AI: Supreme Court overhauls decades of UK case law – Linklaters
- EDPB opinion on AI models: GDPR principles support responsible AI – European Data Protection Board
- Automated decision-making and profiling – Information Commissioner’s Office
- Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank – Courts and Tribunals Judiciary
- Artificial Intelligence (AI) – Judicial Guidance (October 2025) – Courts and Tribunals Judiciary
- Risk Outlook report: the use of artificial intelligence in the legal market – Solicitors Regulation Authority
- Zooming in on AI: tackling deepfakes around the world – A&O Shearman
- New advice for voters on disinformation and campaigners using generative AI – Electoral Commission
- Coalition for Content Provenance and Authenticity (C2PA)
- SynthID – Google DeepMind
- Using Targeted Industrial Policy to Address National Security Implications of Chips – Federation of American Scientists
- How US Export Controls Have (and Haven’t) Curbed Chinese AI – AI Frontiers
- Nvidia and AMD revenue-sharing deal with US government – AP
- US Open-Source AI Governance – Center for AI Policy
- AI Nationalism(s): Global Industrial Policy Approaches to AI – AI Now Institute
- Open Door: AI Innovation in the Global South amid Geostrategic Competition – CSIS
- Energy and AI – International Energy Agency
- China’s East Data West Computing Initiative – Premia Partners
- Measuring and Standardizing AI’s Energy Footprint – Federation of American Scientists
- How AI Threatens Democracy – Journal of Democracy
- Can Democracy Survive the Disruptive Power of AI? – Carnegie Endowment
- AI Index Report 2025 – Stanford HAI
- The Bletchley Declaration – GOV.UK
- India’s AI Impact Summit closes with the New Delhi Declaration – Fortune
- AI Security Institute
- OECD AI Principles – OECD.AI