Who gets to decide the future of AI laws worldwide?

For centuries, legal systems have been constructed on the assumption that technology functions as a straightforward instrument, with responsibility easily assigned to those who wield it. AI, however, defies such logic. Its automation and autonomy extend far beyond familiar boundaries.

“AI is not a simple tool,” Professor Ebers observed. “No one can predict with certainty the outcome when it is deployed. This creates a legal vacuum: if consequences are unclear, how can accountability be enforced?” The advent of generative AI models such as ChatGPT, he added, has only amplified the challenges in devising AI laws.

From the giants of Silicon Valley such as Google, Tesla, Amazon and IBM to ambitious startups, this legal “grey zone” has become a thorny issue. How can AI be rolled out safely without violating laws or endangering users? A single flaw in an algorithm can produce grave consequences, from discriminatory bias to fatal accidents. In 2018, Amazon abandoned an AI recruitment tool after discovering it discriminated against women, having been trained largely on male résumés. Meanwhile, Tesla’s Autopilot system has been linked to 736 crashes in the United States since 2019, with at least 11 fatalities reported by mid-2022. Such incidents have fuelled concerns about over-reliance on AI, while Tesla’s lack of transparency has drawn sharp scrutiny from regulators probing the risks of autonomous systems.

According to the International AI Safety Report presented at the AI Action Summit in Paris in February 2025, risks from generative AI fall broadly into three categories: malicious use (such as fabricating content, swaying public opinion, or aiding cyberattacks); technical failures (including bias, erosion of trust, or loss of control); and systemic risks (ranging from disruption of the labour market to environmental costs and unequal access to global AI research). These concerns are driving governments to design AI laws as legal frameworks to govern the technology.

On 13 March 2024, the European Parliament passed the AI Act, a landmark piece of legislation in the world dedicated to regulating this emerging field. With 523 votes in favour, 46 against and 49 abstentions, the law came into force in May 2024, with implementation phased in from 2025. Roberta Metsola, President of the European Parliament, underlined its significance: “Artificial intelligence has become an important part of our lives. Now, it will also be part of EU law.” Dragos Tudorache, the MEP who led the debate, stressed that the Act is not “an end point” but “a starting point” for a new model of governance that fosters innovation while protecting core values.

Read more: When AI becomes a “colleague”, how should entertainment industry personnel adapt in the digital age?

Drafted in 2021, the AI Act classifies technology into four levels of risk: unacceptable (prohibited outright), high (subject to stringent rules on risk management, data quality, transparency, and cybersecurity), medium, and low. The arrival of ChatGPT in 2022 forced Brussels to add new provisions for general-purpose AI models, whose flexibility makes them difficult to pin down under a single risk category.

Professor Ebers, however, argues that the Act is not a purely risk-based framework. It emphasises risks while giving little weight to benefits, such as the Nobel Prize-winning scientific applications recognised in 2024. Protecting human rights does not sit neatly on a sliding scale of risk: rights are either violated or not, meaning that even minimal risk equates to a breach. Quantifying this remains unresolved. He also criticised the top-down approach, where lawmakers alone decide what is prohibited, high risk, or subject to transparency rules, with limited empirical evidence to justify these classifications. In his view, the EU’s model resembles a political compromise more than a rigorous risk assessment. Moreover, some of the most critical systems such as lethal autonomous weapons or self-driving vehicles, like those implicated in Tesla accidents, remain beyond its scope. Meanwhile, the Act’s sweeping definition of AI, which fails to distinguish between rules-based systems and machine learning, risks penalising technologies that are demonstrably safer. Finally, the requirement to comply with existing laws alongside the new Act creates a double burden for businesses and overlaps in enforcement of AI laws.

The “Brussels Effect”, the tendency for EU rules to shape global markets, may be harder to replicate with AI, according to Professor Ebers. Unlike data privacy, AI spans multiple spheres: from facial recognition to credit scoring, public services to criminal justice. “This requires a national debate: what do we regulate, and how? There is little international consensus,” he remarked. He also highlighted the profound differences between cultures, economies, and political systems.

In Vietnam, AI-generated gambling adverts have recently misused the images, voices and brands of major companies so convincingly that viewers struggle to tell fact from fabrication. At the same time, some organisations have begun producing content with AI without proper oversight, leading to misinformation, doctored images and breaches of sensitive issues. These cases illustrate the pressing need for clear legal guidance to help businesses adopt AI responsibly. Yet Nguyen Quang Dong, Director of the Institute for Policy Studies and Media Development (IPS), stressed that enforcement in Vietnam differs greatly from the EU. Europe benefits from robust courts and supervisory bodies, while Vietnam faces limitations. He suggested that “soft rules”, such as a code of conduct for AI in media and advertising, combined with awareness campaigns and internal risk management, may better suit Vietnam. This approach, he argued, would lower compliance costs for small and medium-sized enterprises while still addressing potential risks.

On 14 June 2025, the Vietnamese National Assembly passed the Law on Digital Technology Industry, which will take effect on 1 January 2026, with some provisions applied earlier from 1 July 2025. Nguyen Khac Lich, Director of the Department of Information Technology Industry, described the move as a historic milestone, positioning Vietnam as a global hub for digital technology. The law provides incentives for projects in AI, semiconductors and data centres, with the ambition of lifting GDP growth by two to three times. It also marks the creation of Vietnam’s first legal corridor for AI, setting principles for development, risk management and oversight of high-risk systems. A controlled testing mechanism (sandbox) has been introduced, enabling companies to pilot new products and encouraging the growth of digital start-ups.

Vietnam’s approach to AI laws draws comparisons with the EU, which separates AI into high-risk and high-impact categories. Yet Professor Ebers warns that such classifications, when not paired with distinct legal obligations, risk creating ambiguity. He also considers the current definition of AI to be overly broad, mirroring international standards, making it difficult to distinguish between systems on the basis of risk. Another concern lies in clarifying responsibilities across the AI value chain, particularly when developers or operators lack insight into the models they are using. The EU has adopted a “transfer of information” mechanism to address this, which Vietnam could adapt. At the same time, the legal framework should underline the benefits of AI to spur innovation, while drawing a clear line between rules-based and machine learning systems so that regulation does not become overbearing. For multi-purpose models such as ChatGPT, responsibilities must be clearly defined according to use case.

The image of Ava walking out of the laboratory unseen in Ex Machina serves as a reminder that AI, left unregulated, can push beyond human boundaries. Laws provide a safeguard, yet there must be equilibrium between restraint and progress. With the Law on Digital Technology Industry, Vietnam is crafting its own framework, informed by European lessons but tailored to local realities. It is not only an exercise in control but also a catalyst for an inventive technological future.

Article adapted from Tatler Vietnam, July 2025 issue

NOW READ

Entertainment industry leaders confront challenges shaping the future

Banned but not beaten: Pita Limjaroenrat is building a new generation of leaders from Harvard

Le Thanh Huong on building Menard Vietnam and empowering women through beauty