Executive Summary
Artificial intelligence (AI) is transforming economies and societies, prompting an urgent need for robust governance frameworks. Governments and international bodies have responded with a patchwork of initiatives — from high-level ethical principles to binding regulations — reflecting both opportunity and concern. For instance, the OECD’s AI Principles (2019, updated in 2024) and UNESCO’s 2021 Recommendation on AI Ethics set global norms for “innovative, trustworthy AI that respects human rights and democratic values”.
In parallel, the EU enacted the world’s first comprehensive AI regulation in 2024 (the “AI Act”), classifying AI systems by risk (prohibited, high-risk, etc.) and extending its rules to any provider targeting EU users. Other countries have adopted varied approaches: the U.S. issued an executive order in 2023 emphasizing AI safety, fairness and worker impacts, while China launched its own AI guidelines on algorithm management and generative AI (e.g. the 2023 “Interim AI Measures” on GenAI).
Despite these efforts, AI governance remains fragmented. We find that global oversight must advance on several fronts: establishing common standards (to avoid a “patchwork” of incompatible rules), ensuring regulatory capacity (especially in developing countries), and addressing new risks (like AI-driven disinformation and job displacement). Critical recommendations include fostering international cooperation (e.g. through OECD, G7/G20, and the UN), updating existing rules in line with AI’s unique challenges, promoting transparency (via algorithm audits and mandatory disclosure), and investing in AI literacy and public awareness. Successfully managing AI will require multi-stakeholder coordination, balancing innovation with protections for human rights, safety, and democratic values.
Introduction
AI systems are now pervasive in commerce, government, and everyday life. Modern AI (including machine learning and generative models) can diagnose diseases, power self-driving cars, streamline logistics, and even generate creative content. At the same time, AI poses novel risks: it can amplify bias, enable mass surveillance or disinformation, and displace workers by automating tasks. These dynamics have spurred policymakers to seek “AI governance” – the set of laws, standards, and institutions that guide how AI is developed and used. This field spans diverse issues: ethical guidelines, data protection, competition, safety, and more. As a result, over 1,000 AI policy initiatives now exist worldwide (covering 69 countries and the EU). However, approaches vary greatly, from voluntary principles to strict controls.
Effective AI governance must keep pace with rapid technology changes. One study notes “AI holds extraordinary potential for both promise and peril”b. On the promising side, AI can boost productivity and solve complex problems (e.g. predicting climate trends). On the perilous side, “irresponsible use could exacerbate societal harms such as fraud, discrimination, bias, and disinformation; [and] displace and disempower workers”. Recognizing this duality, governments aim to harness AI’s benefits while mitigating threats. This report reviews the global landscape of AI regulation and policy, comparing international frameworks and national laws. We analyze key themes – ethics, safety, competitiveness, and inclusion – and conclude with policy recommendations for governments seeking to govern AI in a balanced, forward-looking way.
International and Multilateral Frameworks
Many global bodies have set forth AI principles to guide member countries. The OECD’s AI Principles (first adopted 2019, updated 2024) are “the first intergovernmental standard on AI,” promoting “innovative, trustworthy AI that respects human rights and democratic values”. These include values like human-centeredness, transparency, and accountability. Similarly, UNESCO’s Recommendation on the Ethics of Artificial Intelligence (2021) established a “global standard on AI ethics… applicable to all 194 member states”. UNESCO’s document makes “the protection of human rights and dignity… the cornerstone,” emphasizing principles such as fairness, transparency, and human oversight. It also highlights broad “policy action areas” – data governance, environment, gender, health, and education – to operationalize these principles at national level.
Other international forums have issued high-level commitments. The G7 and G20 statements call for “human-centric, trustworthy AI” and the enforcement of international human rights in AI use. In 2020, 42 countries joined the UN’s “Bletchley Declaration” on AI safety. The Global Partnership on AI (GPAI), OECD’s AI Policy Observatory, and the World Economic Forum’s AI initiatives are examples of multi-stakeholder platforms sharing research and best practices. However, these soft law instruments typically lack enforcement power. They aim instead to build consensus on norms such as privacy protection (echoing GDPR), non-discrimination, and innovation promotion.
Crucially, large economies are applying these principles domestically. For example, the US is developing a voluntary AI Risk Management Framework under NIST, and has issued guidelines on AI-driven FDA approvals in healthcare. The EU’s coordination with neighbours (like through the Council of Europe Convention on AI, Human Rights, Democracy and the Rule of Law) shows steps toward binding commitments. In May 2024, the Council of Europe adopted the first-ever treaty on AI, which commits its 46 members to uphold human rights and democracy in AI deployment. Such developments indicate a slow move from aspirational principles to enforceable standards.
Despite this progress, gaps remain. A recent OECD analysis notes that many countries still lack overarching AI laws: only the EU has a comprehensive AI regulation (enacted 2024). Developing countries, in particular, face capacity constraints. The risk is that without aligned international standards, divergent rules could fragment the digital economy and leave some nations dependent on external tech. For instance, one-third of high-risk AI exposures are concentrated in advanced economies. This underlines the need for cooperative capacity-building (see Recommendations).
National and Regional Regulations
Individual countries have begun enacting concrete rules. The EU’s AI Act (Regulation 2024/1689) is the most ambitious. Adopted in March–July 2024, it classifies AI systems into four risk levels:
- Prohibited AI (e.g. manipulative or social-scoring algorithms) – banned outright.
- High-risk AI (used in critical areas like infrastructure, education, employment, border control, etc.) – allowed only under strict requirements (human oversight, transparency, certified development processes).
- Limited-risk AI (e.g. chatbots, video games) – subject to transparency rules like disclosing automated nature to users.
- Minimal-risk AI – largely unregulated (e.g. spam filters).
The Act also has extraterritorial reach: any AI service targeting EU users must comply, even if developed outside the bloc. This mirrors how the 2016 GDPR extended EU data protection globally. The EU is further working on complementary AI liability rules for harms caused by AI.
In contrast, the United States has not passed an omnibus AI law, but has taken executive and sectoral actions. President Biden’s October 2023 Executive Order directs agencies to ensure AI is “safe and trustworthy” and addresses risks like bias, privacy violations, and labor impacts. The EO instructs agencies to develop guidelines for AI use (e.g. in government procurement), to bolster cybersecurity of AI systems, and to support research.
Some states and cities have imposed local measures (e.g. New York City’s AI hiring tool disclosure law). Legislative proposals have been introduced in Congress on issues like data used for facial recognition or autonomous vehicles, but no consensus has yet emerged. U.S. policy currently emphasizes innovation (e.g. an “AI bill of rights” framework) along with targeted safeguards, leaving substantive enforcement details to federal agencies like NIST, FDA, and FTC.
China has rapidly advanced its own regulatory agenda. In August 2023, it enacted “Interim Measures for Generative AI Services,” the first rule specifically on AI content generation. These rules require GenAI companies to implement content audits and prohibit “immoral” content or disinformation. China’s Cybersecurity Law (2017) and Data Security Law (2021) also cover certain AI uses (notably for personal data and critical information infrastructure). Moreover, government documents have articulated high-level AI ethics principles (e.g. “AI should not go against ethics and morals” as one document put it) and called for controlling recommendation algorithms. Unlike the West, China’s approach is more top-down, linking AI policy to state priorities (economic leadership, “social stability”).
Other nations vary in approach. Japan, South Korea, and Singapore have issued voluntary AI guidelines focusing on industry self-regulation. The United Kingdom plans to rely on existing regulators (e.g. the Information Commissioner’s Office, the Competition and Markets Authority) under a pro-innovation strategy. Some countries (e.g. France, India) emphasize algorithmic auditing or certification programs. In sum, while the EU Act is uniquely comprehensive, most jurisdictions are using a combination of existing laws and new rules to manage AI in key sectors (healthcare, finance, consumer protection, etc.).
Ethical, Security, and Social Challenges
Effective AI governance must grapple with multifaceted challenges. Ethical issues – privacy, bias, and fairness – are foremost. AI systems trained on historic data can perpetuate discrimination (e.g. in hiring or credit). Transparency and explainability are limited for many machine-learning models. Regulatory efforts often include audit, testing, and documentation requirements for high-risk systems to mitigate these concerns (for example, the EU Act’s conformity assessments). But standards for fairness vary: UNESCO advocates that AI uphold cultural diversity and environmental sustainability (reflecting its Human Rights mandate), while China’s standards emphasize social harmony. Aligning these values globally is an ongoing struggle.
Security is another critical dimension. Advanced AI can be dual-use. The recent developments (e.g. ChatGPT, advanced image generation) have spotlighted risks in disinformation, fraud, and cyber-attacks. For instance, generative models are hard to distinguish from human content (“deepfakes”), raising concerns from phishing to political manipulation. A known study warns that sophisticated “compositional deepfakes” could fabricate entire false histories. Recognizing this, the EU Act bans certain manipulative uses of AI outright, and the US EO directs federal agencies to enhance cybersecurity of critical systems against AI-driven threats. The OECD strategic foresight report also emphasizes that AI systems consume significant energy and water, posing environmental challenges that governance must consider.
Economically and socially, AI can amplify inequalities. There is widespread concern about labor markets: OECD analysis finds 27% of jobs are in occupations with high risk of automation (including AI). Recent ILO research shows that about 25% of jobs worldwide could be transformed by generative AI (i.e., change substantially); only a smaller slice (≈3%) are at the highest risk of replacement. Crucially, the impact is uneven: lower-wage occupations face far greater upheaval than higher-skilled jobs, and women may be slightly more affected in certain sectors. If unchecked, AI-driven automation could exacerbate skill gaps and regional disparities (e.g., low-income countries see far less AI exposure to date).
These social challenges have led some regulators to propose redistributive measures. The OECD suggests strengthening safety nets (e.g. unemployment support, retraining funds) for displaced workers. In education, governments are exploring large-scale reskilling: the European Commission, for example, is mobilizing resources to train millions in digital skills by 2030. Ensuring that AI benefits are widely shared requires explicit policy design: an AI economy could, if managed well, generate new high-productivity industries, but without oversight it risks deepening divides.
Altrom’s Policy Recommendations
Given the rapid pace of AI development and its broad impacts, a proactive governance agenda is required. We offer the following recommendations:
- Harmonize Standards Internationally: Countries should converge on core principles and definitions to avoid fragmentation. Multilateral efforts (OECD, UNESCO, G7/G20) should be leveraged to expand adoption of the OECD AI Principles and other guidelines. The Council of Europe Convention (2024) provides a model for binding commitments; endorsing it can help ensure consistent human-rights-based governance across borders. Where possible, regulation (like the EU AI Act) should aim for interoperability or mutual recognition with other regimes. For instance, alignment on “high-risk” categories and standards can ease compliance for global firms while maintaining safety.
- Develop Tailored Regulations with Flexibility: Governments should balance innovation with protection. Following the EU’s risk-based approach, regulators can impose strict rules (human oversight, third-party audits) on applications that directly affect safety or fundamental rights (e.g. autonomous vehicles, medical AI, criminal justice algorithms). Less-sensitive uses can be managed with light-touch oversight, such as transparency requirements. Importantly, regulations need regular review as AI evolves. Establishing expert advisory boards and regulatory sandboxes can allow rules to adapt to new technologies (e.g. updating high-risk lists to include novel generative AI tools).
- Enforce Transparency and Accountability: All stakeholders (companies, governments, researchers) should document and share information on AI systems. This includes requiring providers of high-risk AI to publish data sheets, impact assessments, or risk mitigation plans. We recommend creating mandated audit trails for AI-driven decisions, possibly through independent “AI audit authorities.” For example, some propose that critical AI systems (like automated credit scoring) undergo compulsory algorithmic impact assessments and fair use testing. Enhanced transparency builds public trust and allows regulators to spot issues early.
- Protect Workers and Promote Lifelong Learning: Labor-market policies should accompany AI adoption. This means funding retraining programs in digital and AI-related skills, especially for workers in vulnerable occupations. Public-private partnerships can be formed to anticipate future skill needs (akin to Germany’s vocational system). Educational curricula at all levels should integrate AI literacy. In social policy, consider wage support or earned-income tax credits for those displaced, as the OECD suggests. Where feasible, encourage flexible work arrangements and portable benefits for gig and remote workers, since AI may enable new forms of platform labor.
- Encourage Ethical Research and Innovation: Research funding agencies should require ethics and privacy considerations for AI projects they sponsor. Governments can incentivize “AI for good” initiatives (e.g. grants for AI in healthcare or sustainability) and support open-source AI research to democratize access. At the same time, care is needed for data governance: enforce robust privacy laws (building on models like GDPR) and ensure quality of datasets. Public-sector use of AI must be guided by clear criteria for fairness and non-bias, and citizens should have channels to challenge automated decisions.
- Strengthen Competition and Consumer Protection: AI tends to be data-driven, which can entrench the market power of large tech platforms. Competition authorities should monitor AI markets for monopolistic behavior (e.g. bundling of AI services, data gatekeeping). Consumer protection laws should explicitly cover AI products: regulators may require labels indicating AI-generated content or automated decisions. Considering AI’s cross-border nature, states should collaborate on enforcement (e.g. joint actions against global AI fraud or abuse).
- International Capacity Building: Many developing countries currently lack AI regulatory expertise. International institutions (IMF, World Bank, UNDP) and advanced economies should invest in tech governance capacity-building. This can include sharing best practices, supplying toolkits for AI policy, and financing safe-AI initiatives (e.g. monitoring tools for regional governments). Ensuring that AI governance is inclusive is crucial to prevent a “digital colonization” where only a few countries set the rules.
- Promote Public Engagement and Education: AI regulations often focus on technical aspects, but public attitudes matter. Governments should run outreach to explain both the benefits and risks of AI to citizens, akin to public health campaigns. In particular, schools and universities should integrate AI ethics and critical thinking into curricula. Civil society and media should be empowered to scrutinize AI (e.g. via “algorithmic oversight” journalism). Building societal understanding will help democracies steer AI in alignment with public values.
- Monitor Emerging Threats: Finally, governance mechanisms must keep an eye on new frontiers (like AI in warfare, or synthetic biology aided by AI). National security agencies and international bodies (e.g. UN disarmament committees) need frameworks to assess risks of “extreme AI misuse” and agree on norms (similar to arms control treaties). While such scenarios may seem distant, investment now in horizon-scanning and research could avert future crises.
In summary, Altrom perceives that AI governance is a global challenge requiring coordinated, nimble action. As one independent analyst notes, governance of AI is still “a patchwork [and] evolving field, calling for cooperation among governments, industry and civil society” to ensure AI serves humanity. By embedding AI oversight into existing institutions (justice systems, data regulators, labor ministries) and forging new ones where needed, policymakers can help steer AI towards inclusive growth and social welfare. The principles enshrined by OECD, UNESCO, and others provide a strong ethical foundation; the task now is to translate those principles into practice through effective laws, investment, and global collaboration, which the Altrom Centre highly recommends.
Sources: OECD AI Principles, UNESCO AI Ethics Recommendation, EU Commission and Council of Europe documents, U.S. White House Executive Order, policy analyses (OECD reports, Carnegie Endowment), and international news. (All data and quotes are from cited sources above.)
