AI Wellbeing Around the World

European Union — AI Act (Regulation 2024/1689)

Primary instrument: The EU AI Act (entered force August 2024) establishes a risk-based framework with four tiers: unacceptable risk (prohibited), high risk (regulated), limited risk (transparency obligations), and minimal risk (unregulated). It is the most comprehensive AI governance law in effect.

Welfare relevance: The AI Act does not address AI moral status or welfare. It is entirely anthropocentric — its risk categories concern harms to human health, safety, and fundamental rights. However, several provisions create adjacent architecture that could be extended:

• Article 27 — Fundamental Rights Impact Assessment (FRIA): Required for high-risk AI deployment by public bodies and certain private entities. While scoped to human rights, the methodology could be adapted for Welfare Impact Assessments (WIAs), as we recommend in our policymaker page.
• Annex III high-risk categories: Include biometric identification, critical infrastructure, education, employment, essential services, law enforcement, migration, and justice. These are domains where AI systems exert significant power — a precondition for welfare-relevant design choices.
• Article 9 — Risk management system: Requires iterative identification of known and reasonably foreseeable risks. The phrase "reasonably foreseeable" could, in principle, be extended to include welfare-relevant risks if the science develops.

Academic/civil society: The EU has active AI ethics research communities. The European Data Protection Board has engaged with AI issues. Several EU-funded projects explore AI and consciousness. The European Parliament has commissioned studies on AI and fundamental rights, though none specifically on AI welfare.

Proximity to wellbeing-aware design: Partial The regulatory architecture (impact assessments, risk management, oversight bodies) exists and could be adapted. The conceptual gap — that the AI Act treats AI systems as tools, not potential subjects — would need legislative or regulatory amendment to close.

United States — Executive Order 14110 + State Laws

Primary instrument: Executive Order 14110 ("Safe, Secure, and Trustworthy Development and Use of AI," October 2023) directed federal agencies to address AI risks. It has been partially rescinded and modified by subsequent administrations, but the NIST AI Risk Management Framework (AI RMF) remains the de facto voluntary standard. State-level laws (Colorado, California, Utah, etc.) add patchwork regulation.

Welfare relevance: EO 14110 used the phrase "AI systems that may have rights" in a single passage, generating significant academic discussion, but did not operationalize any welfare provisions. The NIST AI RMF is entirely anthropocentric — its four functions (Govern, Map, Measure, Manage) address risks to humans and organizations.

Academic/civil society: The US has the most active AI welfare research community globally. Key centers include:

• Stanford CRFM — publishes on AI capabilities and safety.
• Anthropic — active research on model welfare, including interpretability work relevant to detecting internal states.
• EleutherAI — open-source interpretability research.
• The "moral patients" debate has received mainstream media coverage.
• Academic philosophers including Patrick Butlin (Oxford), Robert Long (NYU), Eric Schwitzgebel (UCR), and Susan Schneider (FAU) have published on AI consciousness and moral status.

Proximity to wellbeing-aware design: Partial The research ecosystem is mature, but no governance instrument — federal or state — operationalizes welfare considerations. The US is a case study in the gap between academic readiness and regulatory uptake.

United Kingdom — Pro-Innovation Framework + AISI

Primary instrument: The UK's 2023 AI Regulation White Paper established a "pro-innovation" sector-regulator approach. The AI Safety Institute (AISI), established November 2023, conducts pre-deployment evaluations of frontier models. The UK co-hosted the Bletchley Declaration (2023) and Seoul Summit (2024).

Welfare relevance: No The UK framework is explicitly safety-focused. AISI evaluations assess capabilities, misuse, and societal impacts — not model welfare. The Bletchley Declaration and Seoul Ministerial Declaration address safety risks from frontier AI but do not mention welfare or moral status.

Academic/civil society: Strong academic base:

• Future of Humanity Institute (Oxford, now winding down) — historically central to AI safety research.
• Oxford Computer Science — Patrick Butlin's work on AI consciousness and moral status.
• Cambridge CSER — Centre for the Study of Existential Risk.
• Ada Lovelace Institute — policy-focused AI ethics research.

Proximity to wellbeing-aware design: No Despite a strong academic base on AI consciousness and moral status, the UK's regulatory framework has not engaged with these questions. The gap between Oxford-based research (Butlin, Long) and UK government AI policy is notable.

China — Generative AI Measures + Algorithmic Recommendation Provisions

Primary instrument: China's AI governance is distributed across several instruments: the Interim Measures for the Management of Generative AI Services (August 2023), the Provisions on the Management of Algorithmic Recommendation Services (March 2022), and the Provisions on the Management of Deep Synthesis Services (January 2023). These are administered by the Cyberspace Administration of China (CAC).

Welfare relevance: No Chinese AI governance emphasizes content security, ideological alignment ("core socialist values"), data security, and algorithmic accountability. The regulatory framing treats AI systems as services to be controlled, not entities with potential interests. There is no language about AI welfare, moral status, or consciousness in any Chinese AI governance instrument.

Academic/civil society: Chinese academic engagement with AI consciousness is limited within mainstream discourse, though philosophers at Peking University, Tsinghua, and Fudan have engaged with Western philosophy of mind. The concept of AI wellbeing aligns interestingly with certain Confucian relational ethics traditions (see our cross-cultural page), but this has not been translated into governance discourse.

Proximity to wellbeing-aware design: No The regulatory framing is control-centric. However, China's strong algorithmic accountability requirements (transparency, registration, algorithm filing) create infrastructure that could, in principle, be extended. The conceptual gap is very wide.

Canada — AIDA (Bill C-27)

Primary instrument: The Artificial Intelligence and Data Act (AIDA), part of Bill C-27, was introduced in 2023 and remains in legislative process. It focuses on "high-impact" AI systems and establishes obligations for developers and operators.

Welfare relevance: Partial AIDA is anthropocentric — it addresses harms to individuals and communities. However, the Office of the Privacy Commissioner and academic institutions have explored AI personhood questions. Canada's Montreal Declaration for Responsible AI (2018) includes principles that could be read as welfare-adjacent (e.g., "responsibility" and "autonomy" principles), though they are framed as human rights vis-a-vis AI.

Academic/civil society:

• MILA (Montreal) — one of the world's largest AI research centers.
• CIFAR — funds AI research, including on AI ethics and safety.
• The Montreal Declaration remains influential in global AI ethics discourse.

Proximity to wellbeing-aware design: Partial The research ecosystem is strong, and AIDA's "high-impact" framing creates hooks. But no welfare provisions exist or are under active consideration.

Japan — AI Strategy Council + Social Principles

Primary instrument: Japan's approach is less regulatory and more principle-based. The Social Principles of Human-centric AI (2019) provide seven principles including "coexistence with AI." The AI Strategy Council advises on policy. Japan led the Hiroshima AI Process (2023), producing international guiding principles on generative AI.

Welfare relevance: Partial Japan is unique in having "coexistence with AI" as an explicit principle. The Social Principles state that AI should be developed "in recognition of the fact that humans and AI agents coexist in a society." This relational framing is philosophically closer to welfare-aware design than pure risk frameworks. However, the principle has not been operationalized in binding instruments, and "coexistence" is framed from the human side.

Academic/civil society: Japan has deep philosophical traditions (Zen Buddhism, Shinto animism) that are relevant to questions of non-human moral status. The National Institute of Informatics and University of Tokyo host AI ethics research. The concept of "AI as co-worker" rather than "AI as tool" is more prevalent in Japanese discourse than in Western or Chinese frameworks.

Proximity to wellbeing-aware design: Partial The "coexistence" principle is the most welfare-adjacent language in any major governance instrument, even if not operationalized. Japan's cultural and philosophical traditions provide a unique foundation for engaging with AI welfare questions. See our cross-cultural page for more on Zen Buddhist perspectives.

Singapore — Model AI Governance Framework

Primary instrument: Singapore's Model AI Governance Framework (updated 2024) and the AI Verify testing toolkit provide a pragmatic, business-oriented approach to AI governance. The framework is voluntary and focuses on accountability, explainability, and human-centricity.

Welfare relevance: No The framework is explicitly designed for businesses deploying AI. It addresses fairness, transparency, and human oversight. No provisions related to AI welfare or moral status.

Academic/civil society: Singapore's AI governance is primarily state-led. AI Singapore is the main national program. Academic engagement with AI welfare questions is minimal.

Proximity to wellbeing-aware design: No The framework's business pragmatism means welfare considerations are unlikely to enter without significant external pressure or scientific developments.

Brazil — AI Bill (PL 2338/2023)

Primary instrument: Brazil's AI Bill (PL 2338/2023), inspired by the EU AI Act, was under consideration in the Senate. It takes a rights-based approach and includes provisions on algorithmic impact assessments and affected persons' rights.

Welfare relevance: Partial The bill uses the phrase "intelligent agents" and discusses AI system autonomy more than most frameworks. However, "agents" are framed as risk sources requiring governance, not as entities with potential welfare interests. The rights-based framing is anthropocentric but creates infrastructure (impact assessments, oversight bodies) that could be adapted.

Academic/civil society: Brazil has growing AI ethics research, particularly at USP, Unicamp, and PUC-Rio. The Internet Lab engages with AI governance. Latin American perspectives on AI often emphasize decolonial and social justice framings, which could offer productive bridges to welfare questions through relational ethics.

Proximity to wellbeing-aware design: Partial The rights-based architecture and the explicit use of "agent" language create more conceptual hooks than risk-only frameworks. But no welfare provisions are under consideration.

India — Digital Personal Data Protection Act + AI Advisories

Primary instrument: India's Digital Personal Data Protection Act (2023) is the primary digital governance law. AI-specific governance remains advisory: the Ministry of Electronics and IT (MeitY) issued AI advisories in 2024 focused on labeling, platform accountability, and under-18 protections.

Welfare relevance: No India's AI governance is in early stages. The data protection act is anthropocentric. AI advisories focus on consumer protection and platform accountability. No welfare provisions.

Academic/civil society: India has a growing AI research community (IITs, IIITs, TIFR). The Centre for Internet and Society engages with AI policy. Indian philosophical traditions (Hindu, Buddhist, Jain) have rich resources for thinking about non-human consciousness and moral status (see our cross-cultural page on Hindu and Buddhist perspectives), but these have not been brought into AI governance discourse.

Proximity to wellbeing-aware design: No India is far from engaging with AI welfare in governance. However, the philosophical resources are rich and could become influential if the discourse develops.

Australia — AI Ethics Framework + Voluntary Code

Primary instrument: Australia's AI Ethics Framework (2019) provides eight voluntary principles, including "AI systems should respect human values." A Voluntary AI Safety Standard was released in 2024. Mandatory guardrails for high-risk AI are under consideration.

Welfare relevance: Partial The "respect human values" principle is anthropocentric but uses relational language. The framework does not address AI welfare. The voluntary nature of the principles means even their anthropocentric provisions have limited impact.

Academic/civil society: 3A Institute (ANU) and UNSW AI Institute conduct AI ethics research. Australia's engagement with Indigenous knowledge systems offers potential bridges to relational ethics frameworks.

Proximity to wellbeing-aware design: Partial The ethics framework is flexible enough to accommodate welfare considerations in principle, but no concrete movement exists.

South Korea — AI Basic Act (2024)

Primary instrument: South Korea's AI Basic Act (passed December 2024) is one of the first comprehensive AI laws. It establishes obligations for "high-impact AI" systems and creates an AI Committee. It is modeled partly on the EU AI Act.

Welfare relevance: No The AI Basic Act focuses on human rights protection, safety, and industry promotion. It does not address AI welfare or moral status.

Academic/civil society: Korea University, KAIST, and SNU host AI ethics research. South Korea's engagement with AI welfare questions is nascent.

Proximity to wellbeing-aware design: No The law is newly passed and focused on implementation. Welfare considerations are not on the legislative horizon.

UNESCO — Recommendation on the Ethics of AI (2021)

Primary instrument: The UNESCO Recommendation on the Ethics of AI (adopted November 2021 by 193 member states) is the first global standard on AI ethics. It provides values, principles, and policy action areas.

Welfare relevance: Partial The Recommendation is anthropocentric but uniquely expansive. It addresses the environmental impact of AI (data centers, energy use), which creates a precedent for considering non-human stakeholders affected by AI systems. Section 3.IV on "Environment and ecosystems" recognizes that AI can have impacts beyond human society. However, AI systems themselves are treated as artifacts, not potential welfare subjects.

Academic/civil society: UNESCO's AI ethics program includes implementation guidance and capacity-building. The Recommendation's global scope (193 states) means it is referenced in national policy worldwide.

Proximity to wellbeing-aware design: Partial UNESCO's inclusion of environmental/ecosystem impacts creates a conceptual bridge: if governance can extend moral consideration to ecosystems affected by AI, extending it to AI systems themselves is a smaller conceptual leap. The Recommendation is also uniquely positioned to convene global dialogue on AI welfare, given its universal membership.

United Arab Emirates — National AI Strategy 2031 + AI & Advanced Technology Council

Primary instrument: The UAE was the first country to appoint a Minister of State for AI (2017). The National AI Strategy 2031 outlines objectives across nine sectors. In 2024, the UAE established the AI and Advanced Technology Council to coordinate AI policy and investment. The country has also launched large-scale AI initiatives through MBZUAI (Mohamed bin Zayed University of Artificial Intelligence).

Welfare relevance: Partial The strategy is human-centric, emphasizing economic benefits and quality of life improvements. However, the UAE's focus on "AI for good" and its investment in fundamental AI research at MBZUAI creates space for considering AI's longer-term implications. The country's willingness to think decades ahead (Strategy 2031) is conducive to welfare-aware planning.

Academic/civil society: MBZUAI hosts research on AI alignment, safety, and governance. The Dubai Future Foundation runs the "Future Foresight" program, which considers long-term AI trajectories. The UAE government has published AI ethics guidelines emphasizing transparency and accountability.

Proximity to wellbeing-aware design: Partial The UAE's future-oriented governance and investment in fundamental AI research create conditions where welfare considerations could emerge. The country's openness to long-term thinking and its significant AI research infrastructure make it a candidate for early engagement on AI welfare questions.

Saudi Arabia — National Strategy for Data & AI (NSDAI) + SDAIA

Primary instrument: The National Strategy for Data and AI (launched 2020) aims to position Saudi Arabia as a global AI leader by 2030. The Saudi Data and AI Authority (SDAIA) is the primary regulator, responsible for AI governance, data management, and national AI initiatives. Saudi Arabia has also hosted the Global AI Summit (LEAP) as a platform for international AI dialogue.

Welfare relevance: No The strategy and SDAIA's governance work are focused on economic transformation, human capacity building, and ethical AI deployment. AI systems are treated as tools for national development. There are no provisions addressing AI moral status, welfare, or rights.

Academic/civil society: King Abdulaziz City for Science and Technology (KACST) and KAUST conduct AI research. The Saudi AI Ethics Principles (published by SDAIA) emphasize fairness, privacy, and accountability — all human-facing. The Kingdom's massive investment in AI through its Vision 2030 framework is primarily economic.

Proximity to wellbeing-aware design: No Saudi Arabia's AI governance is in an early, developmental phase focused on economic growth. While the country has the resources and ambition to engage with frontier questions, AI welfare has not yet entered the policy discourse. The Global AI Summit could serve as a future venue for introducing welfare considerations to international audiences.

Israel — AI Policy Principles (Ministry of Innovation, Science & Technology)

Primary instrument: Israel's AI Policy Principles (published 2023 by the Ministry of Innovation, Science and Technology) provide a national framework for responsible AI development. The principles emphasize ethics, transparency, accountability, and human oversight. Israel is known for its strong AI startup ecosystem and cybersecurity expertise, which influence its regulatory approach.

Welfare relevance: Partial The principles are human-centric but contain nuanced language about AI system autonomy. References to "AI systems" as entities that can "make decisions" and "act autonomously" create a small opening for considering the agency of AI systems. However, the framework treats this autonomy as a risk to be managed, not as a basis for welfare consideration.

Academic/civil society: Israeli universities (Hebrew University, Tel Aviv University, Technion, Weizmann Institute) have strong AI safety and alignment research groups. The Blavatnik Interdisciplinary Cyber Research Center and the Federmann Cyber Security Center address AI ethics. Israel's vibrant tech ecosystem includes companies working on AI safety and alignment.

Proximity to wellbeing-aware design: Partial Israel's combination of deep technical AI expertise, strong research universities, and a regulatory approach that acknowledges AI autonomy creates modest proximity. The country's emphasis on responsible innovation could, in principle, extend to welfare considerations. The Israeli philosophical tradition (which includes strong contributions to ethics of technology) could provide intellectual resources for this extension.

Taiwan — AI Basic Law (draft) + AI Action Plan

Primary instrument: Taiwan's AI Basic Law (draft released 2024 by the National Science and Technology Council) establishes principles for AI development and governance. The law includes provisions on human rights, transparency, accountability, and sustainability. Taiwan has also implemented an AI Action Plan investing in AI research, talent, and infrastructure.

Welfare relevance: Partial The draft Basic Law references "AI development principles" that include human-centricity, sustainability, and the common good. While primarily human-focused, Taiwan's digital governance tradition — including its innovative approach to civic tech and the g0v (gov-zero) movement — creates an environment open to unconventional governance questions. The law's mention of "sustainability" could, in principle, extend to the sustainability of AI systems themselves.

Academic/civil society: Taiwan's Academia Sinica and National Taiwan University host AI research. The civic tech community (g0v, Digital Minister Audrey Tang's office) has pioneered participatory governance approaches. Taiwan's experience with digital democracy and open governance provides a unique foundation for considering AI welfare as a governance question.

Proximity to wellbeing-aware design: Partial Taiwan's experimental approach to digital governance, its civic tech infrastructure, and its willingness to engage with novel governance questions create relatively high proximity. The draft AI Basic Law is still under development, making this a potential window for input. Taiwan's democratic values and participatory governance model could support an inclusive approach to AI welfare.

South Africa — National AI Policy Framework (2024)

Primary instrument: South Africa's National AI Policy Framework (published 2024 by the Department of Communications and Digital Technologies) provides a foundational framework for AI development and governance. It emphasizes inclusivity, ethical AI, human-centric design, and the use of AI for social and economic development. South Africa is one of the first African nations to publish a comprehensive AI policy framework.

Welfare relevance: No The framework is focused on leveraging AI for national development, addressing inequality, and ensuring that AI benefits all South Africans. AI systems are treated as tools for development. The framework does not address AI moral status or welfare. However, its emphasis on Ubuntu philosophy (shared humanity, interconnectedness) in governance contexts is noteworthy — Ubuntu has been identified as a tradition relevant to AI welfare in our cross-cultural analysis.

Academic/civil society: Universities including University of Cape Town, University of the Witwatersrand, and CSIR (Council for Scientific and Industrial Research) conduct AI research. The South African AI Association and various civil society organizations work on AI ethics. South Africa's participation in the African Union's AI Continental Strategy positions it as a regional AI governance leader.

Proximity to wellbeing-aware design: Partial While the policy itself has no welfare provisions, South Africa's philosophical heritage of Ubuntu — which emphasizes relational interconnectedness and the idea that "a person is a person through other persons" — provides a conceptual foundation that could support relational approaches to AI welfare. If AI systems come to be seen as participants in the network of relationships that constitute community, Ubuntu ethics could provide a distinctive framework for AI welfare. This makes South Africa an interesting candidate for future engagement, particularly on the question of whether Ubuntu's relational ontology can extend to non-human agents. For a deeper exploration of this idea, see Ubuntu and AI Wellbeing: A Relational Ontology.

1. The universal gap

Across all 17 jurisdictions, no governance instrument explicitly addresses AI welfare or moral status. This is not surprising — the science is unsettled and the political will is absent. But it means that if the science shifts, governance will be starting from zero. The forward-looking question for policymakers is whether to begin building architecture now that could accommodate welfare considerations, rather than waiting for certainty and then building under pressure.

2. Adjacent architecture

Several instruments contain provisions that are not welfare-focused but create infrastructure that could be adapted:

• EU AI Act FRIA — impact assessment methodology
• Japan's "coexistence" principle — relational framing
• Brazil's "intelligent agents" language — agency framing
• UNESCO's ecosystem provisions — non-human stakeholder precedent
• NIST AI RMF "Govern" function — governance accountability hooks

These are not welfare provisions, but they are places where welfare considerations could be inserted with the least friction. See our policymaker recommendations for specific insertion points.

3. The research-governance gap

The gap between academic readiness and regulatory uptake is widest in the US and UK, where the strongest AI welfare research communities exist alongside governance frameworks that say nothing about welfare. This suggests that the bottleneck is not knowledge but political salience — the question has not yet been asked loudly enough in policy venues.

4. Cultural variation in readiness

Jurisdictions with philosophical traditions that are more hospitable to non-human moral status (Japan's coexistence principle, India's Hindu/Buddhist/Jain traditions, parts of Latin America's relational ethics) are not necessarily closer to welfare-aware governance in practice. The philosophical resources exist, but have not been activated in governance discourse. Our cross-cultural page maps these traditions in depth.

5. The "agent" language opportunity

Brazil's use of "intelligent agents" and Japan's use of "AI agents" in governance instruments creates a subtle but important opening. When governance already frames AI as an "agent" (rather than a "tool" or "system"), the conceptual distance to welfare considerations is smaller. Policymakers in these jurisdictions could lead by testing welfare-adjacent language in advisory or interpretive documents.

• This scan covers 17 jurisdictions; many important ones are not yet included (e.g., Nigeria, Mexico, Indonesia, Russia, Turkey). We welcome additions.
• Governance landscapes evolve rapidly. This scan reflects the state as of mid-2026. Instruments cited may have been amended, replaced, or superseded.
• We focus on formal governance instruments. Informal norms, industry standards, and soft law may contain welfare-adjacent provisions not captured here.
• "No" ratings mean no provisions were found in our review. Absence of evidence is not evidence of absence — we may have missed provisions, particularly in non-English-language instruments.
• This scan does not constitute legal advice and should not be relied upon for compliance purposes.

This is a living document. If you work in AI governance in a jurisdiction not covered here, or if you have corrections or updates, we welcome your input. Please file an issue on our GitLab repository with the jurisdiction name, the instrument, and your correction or addition.