As AI rapidly transforms industries, economies, and societies, governments and businesses worldwide are grappling with how to regulate its development and deployment.

The stakes couldn’t really be higher and the enterprise world seems to be all in – AI promises immense benefits, from revolutionising healthcare to optimising supply chains, but it also poses significant risks, including bias, privacy violations, and even threats to democratic processes.

Seeking to give a comprehensive overview, Arm is releasing a five-phase report designed to reflect organisations’ current state of readiness for successful AI deployments and serve as a guide for leaders on future direction.

The AI Readiness Index, a comprehensive global study surveying over 650 business leaders, sheds light on the current state of AI governance. The third installment, titled Policy and Governance: Shaping the AI Regulatory Landscape, explores the multifaceted challenges of regulating AI – a technology that defies easy categorisation due to its diverse applications, from chatbots to autonomous weapons.

Below, we’ll take a look at some of the reports key insights.

The Complexity of AI Governance

AI governance is inherently complex, shaped by a patchwork of local, national, and international regulations. Unlike more traditional technologies, AI spans multiple domains – foundation models (like large language models), AI-powered physical products (such as medical devices), small-scale AI services, and military AI applications – each requiring distinct regulatory approaches.

For example, foundation models, trained on vast datasets and capable of performing a wide range of tasks, present unique risks. Their potential misuse – such as generating deepfakes or automating disinformation campaigns – has led to calls for preemptive safety measures, including international agreements to prevent malicious development.

Meanwhile, AI in physical products, from smart appliances to autonomous vehicles, demands stringent safety and cybersecurity standards to prevent real-world harm.

Military applications of AI raise even thornier ethical questions, particularly around autonomous weapons systems and the role of human oversight in life-or-death decisions.

The lack of global consensus on these issues underscores the difficulty of creating cohesive governance frameworks.

Countries are taking vastly different approaches to AI regulation, reflecting their political, economic, and ethical priorities.

Below, we’ll take a look at how the report breaks down some of the big players on the global stage.

The EU’s Risk-Based Model

In August 2024, the European Union enacted the AI Act, the world’s first comprehensive AI law. The legislation adopts a risk-based approach, categorising AI systems into three tiers:

• Unacceptable risk (e.g., social scoring, real-time biometric surveillance) – banned outright.
• High risk (e.g., AI used in hiring, critical infrastructure) – subject to strict transparency and compliance checks.
• Limited risk (e.g., chatbots, recommendation algorithms) – minimal transparency requirements.

The EU’s stringent rules aim to prioritise human rights and accountability, but critics argue they could stifle innovation. Some companies, like Meta, have already pulled AI services from the EU market, citing regulatory uncertainty.

The U.S.’s Light-Touch Strategy

Unlike the EU, the United States lacks a unified federal AI law, instead relying on sector-specific regulations (e.g., healthcare, finance) and voluntary industry guidelines. The National AI Initiative Act of 2020 promotes AI development but imposes few binding rules.

This hands-off approach has drawn criticism for allowing potential harms – such as AI-driven hiring bias or deepfake exploitation – to go unchecked. However, proponents argue it fosters innovation and flexibility, keeping the U.S. competitive in the global AI race.

China’s State-Controlled AI Development

China’s AI governance prioritises state control and economic growth, with regulations like the New Generation AI Development Plan and Algorithmic Regulation Provisions ensuring alignment with Communist Party objectives. While China encourages rapid AI advancement, it imposes strict oversight on facial recognition, deepfakes, and autonomous vehicles to maintain what the CCP deems as social stability.

Singapore’s Balanced Framework

Singapore’s Model AI Governance Framework emphasises transparency, fairness, and accountability without heavy-handed regulation. The government encourages self-assessment tools like AI Verify, allowing businesses to test AI systems responsibly while avoiding excessive compliance burdens.

The Challenge of Global Coordination

With AI systems transcending borders – often accessible via a simple VPN – national regulations alone are insufficient. The report states that international cooperation is critical to prevent regulatory arbitrage, where companies exploit lax jurisdictions to deploy risky AI.

Efforts like the OECD’s AI Principles and UNESCO’s Ethical AI Recommendations provide foundational guidelines, but geopolitical tensions hinder deeper alignment. The EU, U.S., and UK are collaborating on foundation model governance, while China remains at odds over data privacy and state surveillance.

Beyond legal compliance, ethical concerns loom large. AI systems can perpetuate bias (e.g., discriminatory hiring algorithms), violate privacy (e.g., unauthorised data scraping), and operate as “black boxes” with unexplainable decision-making processes.

To address these risks, organisations are adopting tools like:

• ISO/IEC 42001 – A global standard for AI management systems.
• NIST’s AI Risk Management Framework – A voluntary U.S. guideline for ethical AI development.
• Automated red-teaming – Simulating cyber-attacks to identify AI vulnerabilities.

Private companies also play a crucial role, with some establishing internal ethics boards to monitor AI deployments. Yet, without enforceable global standards, ethical AI remains a voluntary pursuit for many.

The Future of AI Regulation: Adaptive Policies and Innovation Incentives

As AI evolves, regulators face a dilemma: how to enforce safety without stifling progress? Proposed solutions in the report include:

• Regulatory sandboxes – Controlled environments for testing AI innovations.
• Dynamic legislation – Laws that adapt to technological advancements (e.g., the EU’s updatable AI Act).
• Content watermarking – Identifying AI-generated media to combat misinformation.

Meanwhile, governments are investing heavily in AI infrastructure, recognising that compute power is as critical as software development. The U.S., EU, and Asia are racing to build the hardware needed to sustain AI leadership.

Though, it should be noted that the report doesn’t mention the immense toll that AI is taking on data centres – recent data from Ciena found that increasing energy and data demands of AI means that nearly half (44%) of all new UK data centres are predicted be dedicated for AI workloads over the coming five years.

Beyond this, another recent report found that almost three-quarters (74%) of data centre IT, facility, and business decision makers report being under increased pressure from AI-driven demands. 

Recommended reading

• AI Adoption Stumbles as Report Highlights Employee-leadership Disconnect
• 8 in 10 Hiring Managers Prioritise AI Skills When Hiring
• Is AI Now Essential in Recruitment?

AI’s transformative potential is undeniable, but so are its risks.

This latest chapter in Arm’s ambitious report series shows just how fragmented the global regulatory landscape remains – the EU prioritising safety, the U.S. favoring innovation, and China leveraging AI for state control, in a nutshell.

For businesses, navigating this patchwork requires agility, foresight and hopefully, a bit of compassion – complying with strict EU rules while adapting to America’s sector-specific approach and China’s political mandates. International collaboration, ethical safeguards, and adaptive policies will be key to ensuring AI benefits society without compromising rights or security.