Terms and Conditions Apply (To Everything Except the Algorithm): Who Regulates the Regulators of the Machines that Regulate the Market?
By Isabelle Teo Zi Ning, mentored by Vimaljit Kaur and Dominic Tayco
I. Introduction
A. The purpose behind legislation
1. A commonly held assumption is that a society’s constitution incorporates its values into its legislation and statutes.[1] Indeed, this seems to be the ideal. Article 152 of Singapore’s Constitution records the country's commitment to its racial and religious minorities, but also making it the responsibility of the Government "constantly to care for the interests" of those minorities, which is a duty of continuing performance rather than a settled guarantee.[2] Article 152(2) provides for an outcome to be pursued rather than a procedure to be followed. Beyond having the Constitution supply the machinery for discharging it, through the Presidential Council for Minority Rights and through parliamentary representation designed to return minority members,[3] successive governments have given effect to it through ordinary policy thorugh educational and community institutions established by statute, and self-help organisations operating under state sponsorship.[4] Rather than saying a constitution automatically carries a society’s values, a more apt framing might be that it is a promissory note that society must continually redeem.
2. However, beyond statutes, the common law allows judges to read and interpret legislation to apply it in situations where a plain reading of legislation would be insufficient to reach a solution.[5] This gap-filling exercise is also another reading of society’s values, where a judge interpreting legislation against a novel harm must reach for the normative rules of society.
3. Legislation is manufactured by governments.[6] However, where governments are a function of society, what they manufacture is compiled from a rubric of customs, norms, and expectations preceding every drafting instruction.[7]
4. Notably, these delineations of societal values, be it in legislation or in judicial case law, are subject to the existence of these legislative discussions and judicial hearings. Generally, this requires there to be an issue with the current state of affairs, for someone to bring It up, and for there to be discussions of a solution. However, artificial intelligence (“AI”) iterates in months, leading to another issue—whether legislation can catch up with the developments in AI.[8]
5. This is a fifty-year-old dilemma, where David Collingridge observed in 1980 that the social control of technology faces two issues simultaneously—early on, when a technology is still malleable, we lack the knowledge of what harms need preventing; by the time the harms are understood, the technology is so entrenched that control is slow, costly and disruptive.[9] Larry Downes later recast the same tension as the “pacing problem”, whereby technology advances exponentially while law, economics and social norms adapt only incrementally.[10] AI elevates dilemma to a level no prior technology has matched, because its generations now turn over faster than legislative cycles.[11] The instruments available run on a spectrum from hard law which is binding,[12] to soft law, which includes the “variety of non-binding norms and techniques” that create expectations without direct enforceability.[13] The structural pressure of the pacing problem, as Hagemann, Huddleston, and Thierer argue, pushes steadily towards the soft end, because formal legislation simply cannot move at the speed of the developments.[14]
6. Thus, when it comes to AI, the two main paths that every legal system appears to deliberate between are that of reformation or adaptation. If a legal system chooses adaptation, this would look like taking a generous reading of the relevant statutes, stretching consumer, privacy, and product law over the new technology, and legislating only where a specific gap is proven.[15] Alternatively, if a legal system chooses reformation, this requires an entire audit of the present technology laws, before enacting a new framework that covers what is now visible and what is expected to come.[16]
B. Thesis and current literature
7. The choice taken up by different legal systems will depend on the legal system’s culture and what they hope to achieve. Thus, this essay argues that the choice of how different regimes choose to govern AI is a reflection of the trust a society has in it, which is evident in the society’s background and values.
8. Most current literature treats AI regulation as an instrument of foreign policy, written in the shadow of a Sino-American duopoly.[17] Even in evaluating these national statutes, they are usually measured against the EU AI Act.[18] Scholarly literature may also analyse these instruments as against the nation’s position in the AI race.[19]
9. However, before a law is a move in a geopolitical game, or a better or worse copy of someone else’s, it is a promise a community makes to itself about what it will protect and what it will permit.[20] We cannot ask whether an AI statute is “good” without first asking whose expectations it was built to meet.[21]
10. This article will thus analyse the choices as between Singapore and Australia. These are both children of the English common law with similar legal forms. They are received through the same imperial history, operating through Westminster-descended parliaments, and construed by courts that cite one another.[22] Thus, whatever divergence found between them cannot be explained by legal form, and the comparison therefore isolates each society’s values to better understand what separates these two jurisdictions’ choices, which embodies the thesis of this essay.
II. Singapore
A. Singapore society
11. Post-war Singapore had to build its collective from scratch.[23] Singapore’s founding father, Lee Kuan Yew, installed survival as a national creed, colouring Singapore as a “little red dot” of no resources except its people.[24]
12. Modern Singapore’s entire founding is built on her vulnerability.[25] Thus, everything protective is derived from this vulnerability. This leads to peace being deliberately engineered through legislation and policies, including the Maintenance of Religious Harmony Act, which arms the state with pre-emptive restraining orders against those who stir inter-religious enmity.[26] Integration is thus a designed outcome as opposed to being left to markets, extended as recently as 2025 with a new racial-harmony statute.[27]
13. This is also codified in the 1991 Shared Values White Paper, which ranks national values in order: “Nation before community and society above self; Family as the basic unit of society; Community support and respect for the individual; Consensus, not conflict; Racial and religious harmony”.[28] The individual appears only after nation and family and is further caveated by consensus and harmony.[29]
14. This also carries on into how Singapore approaches technology. The first is pragmatism, where the governing test, in Lee Kuan Yew’s own words, was “does it work?” — an outcome-based legitimacy that judges a law by its results rather than its doctrine.[30] The second is total and communal defence.[31] Since 1984, Singapore has different pillars of security that is brought to the attention of Singaporeans annually.[32]Notably, digital defence was introduced as a pillar in February 2019, formally making cyber hygiene and resistance to online falsehood a civic duty on par with military service.[33] This explicated Singapore’s acknowledgement of the possible impact the digital world could have on the physical world, especially if not properly maintained.[34]
15. Indeed, this concern proved true as with the 2021 TraceTogether incident when it emerged in January 2021 that police could access TraceTogether contact-tracing data after ministers had extensively promised it would be used only for contact tracing.[35] The backlash was immediate, such that Parliament passed emergency legislation within a month to ring-fence access to seven categories of serious crime, later winding the system down and deleting the data.[36]
16. The digital society produced is state-mediated and thus, it is unsurprising Singapore takes an interventionistic stance towards building its digital society. This is further legitimised by how Singaporeans’ trust in government runs around 76%, far above global averages.[37]
B. Singapore’s regulation
17. Singapore governs AI through a voluntary, framework-driven model.[38] There is no specific legislation directly regulative AI.[39] Instead, Singapore relies on a combination of national strategies, voluntary governance frameworks, sector-specific guidance, and practical implementation tools.[40] Further, where the government has consistently signalled it does not intend to enact comprehensive horizontal AI legislation.[41] The Minister for Digital Development has described the approach as a practical one, leveraging existing laws and regulations wherever possible, while remaining “ready to add to the existing suite” when necessary.[42] This supports its pro-innovation stance, striking a balance between allowing AI innovation to develop and safeguarding the public interest in AI ethics and governance.[43]
18. Given the non-binding nature of Singapore’s approach, it would make sense that there are no proper legal definition of AI and no formal or statutory risk tier. Rather, Singapore’s framework is intentionally distributed across multiple public authorities and is undergirded by institutional guidance.[44]
19. The centrepiece is the Model AI Governance Framework, now in its third generation.[45] The frameworks translate principles into operational measures.[46] While voluntary, they are increasingly treated as de facto best practice.[47] Each version focuses on different aspects of AI, focusing on the following:
2019 (Model Artificial Intelligence Governance Framework) | 2024 (Model AI Governance Framework for Generating AI—Fostering a Trusted Ecosystem) | 2026 (Model AI Governance Framework for Agentic AI) |
- Internal governance; - Decision-making models; - Operations management; and - Customer relationship management. | - Accountability; - Data; - Trusted development and deployment of AI; - Incident reporting; - Testing and assurance; - Security; - Content provenance; - Safety and alignment research and development; and - AI for the public good. | - Assess and bound the risks upfront; - Make humans meaningfully accountable; - Implement technical controls and processes; and - Enable end-user responsibility. |
20. The evolution of the various frameworks demonstrates the government’s incorporation of industry feedback, which especially reflect the pace at which the industry is moving.[48]
21. The first edition, released in January 2019 and revised in January 2020, translated ethical principles into operational measures across internal governance, the level of human involvement in AI-augmented decision-making, operations management, and stakeholder interaction.[49] The Model AI Governance Framework for Generative AI followed on 30 May 2024, organised around nine dimensions spanning accountability, data, trusted development and deployment, incident reporting, testing and assurance, security, content provenance, safety and alignment research, and AI for the public good.[50] A Model AI Governance Framework for Agentic AI was launched on 22 January 2026 and has already been revised twice within months, directing organisations to assess and bound risks upfront, keep humans meaningfully accountable, implement technical controls across the agent lifecycle, and enable end-user responsibility.[51]
22. Overall, there are a few recurring themes across all three versions of the Model AI Governance Framework, including:
(a) Holding the relevant stakeholders accountable;
(b) Security risks (and knowing of them); and
(c) Growing the AI scene in Singapore.[52]
23. The frameworks are supported by testing infrastructure.[53] AI Verify, launched as a pilot in May 2022 as the world’s first AI governance testing framework and software toolkit, allows organisations to validate the claimed performance of their AI systems against the framework’s principles through standardised tests.[54] It was open-sourced under the AI Verify Foundation in June 2023.[55] The deliberate interoperability of Singapore’s instruments with international standards (including the United States NIST AI Risk Management Framework) reduces the compliance friction that would otherwise make a regional hub less viable for multinationals.[56]
24. Further, an apex political body has been added.[57] Announced in the Budget 2026, the Prime Minister announced a National AI Council, set to be chaired by himself.[58] It specifically focuses on the areas of advanced manufacturing, connectivity, finance, and [ITZN1] healthcare.[59] This council is meant to provide the necessary strategic direction and oversee the development and execution of AI missions to unlock regulations and resources to accelerate the development, testing, deployment, and scaling of AI solutions.[60]
25. Beyond non-binding frameworks, certain sector-specific rules carry the operative weight. For example, health authorities issued Artificial Intelligence in Healthcare Guidelines in 2021, updated in March 2026.[61] Further, the Personal Data Protection Commission issued advisory guidelines on the use of personal data in AI systems in March 2024.[62] Hard law exists only where a specific harm has crystallised, such as how the Elections (Integrity of Online Advertising) (Amendment) Act 2024 prohibits digitally manipulated online election advertising that realistically misrepresents a candidate.[63] Notably, this was only passed within months of the harm becoming salient.[64]
26. Lastly, the model is exported.[65] Singapore championed the ASEAN Guide on AI Governance and Ethics, endorsed under its chairmanship of the ASEAN Digital Ministers’ Meeting in February 2024 and expanded to cover generative AI in January 2025, and established the ASEAN working group on AI governance.[66] The more Singapore’s governance model is adopted as a regional benchmark, the more its position as neutral, trusted infrastructure becomes self-reinforcing.
III. Australia
A. Australian society
27. Australian society is amongst the most migrant-composed, where more than half of Australians were born overseas or who have a parent who was born overseas, and Australians are amongst the most secular.[67] Thus, society would be more concerned about what governmental institutions say, rather than any shared ideology.
28. Ironically, Australia’s legal culture involves adversarial enforcement.[68] Outside the United States, Australia is “perhaps the preferred jurisdiction to file a class action”, where civil society is unafraid to file court proceedings.[69] This means that while Australia has a high tolerance for statutory intervention, Australians are also highly suspicious of unearned authority.[70]
29. Australia has a substantial but not exceptional level of AI adoption, with half of Australians reporting regular or semi-regular intentional use of AI, yet only 30% of Australians consider that the benefits of AI outweigh its risks.[71] This reluctance is commonly attributed to an unlawful automated welfare debt recovery system run by the Australian government from 2016 to 2019 called Robodebt, which accused 470,000 people of owing money from its own internal algorithm.[72] Not only did this lead to a record quantum of a single legal settlement, but also led to a huge decrease in institutional trust, where Australians demanded individual, human accountability.[73]
30. To balance this demand against Australia’s scepticism, and indeed in line with its adversarial nature, Australia assertively regulates technology. The digital society this produces is court-mediated and thus, it is unsurprising Australia takes an adversarial stance towards governing its digital society.[74]
B. Australian regulation
31. Australia has declined to enact an AI statute, instead choosing to rely on borrowed instruments.[75] While Australia has not legislated at AI, it has legislated around AI through hard statutes whenever a concrete harm has crystallised.[76]
32. However, the trajectory that produced this position was not linear. Australia issued the following frameworks:
(a) In 2019, eight voluntary AI Ethics Principles in 2019;[77]
(b) in September 2024, the Department of Industry, Science and Resources published a proposals paper canvassing ten mandatory guardrails for AI in high-risk settings, expressly contemplating a whole-of-economy AI Act with a dedicated regulator;[78] and
(c) in December 2025, the National AI Plan abandoned that proposal altogether, confirming that existing technology-neutral laws and sectoral regulators would remain the foundation.[79]
33. Yet, Australia shifted its position again on 15 July 2026, where the Prime Minister announced a set of Australian Standards for AI carrying mandatory obligations, established an Office of AI within the Department of the Prime Minister and Cabinet, and committed to introducing legislation in early 2027 following a National Cabinet process in August 2026.[80]
34. Notably, these attach to large data centres and to the use of Australian creative works in training.[81]Substantively, this means Australia is utilising hard law and legislating at the infrastructure and input layers, of which an extraterritorial non-producer retains practical control.
35. However, the Australian standards do not govern how an organization deploys an AI system, so deployment would refer back to the previous “layers” that Australia has already put out.[82] These include the National AI Plan of 2 December 2025 sets a direction rather than a duty, confirming that existing laws and sectoral regulators will remain the foundation, supported by voluntary guidance and a new AI Safety Institute, and expressly declining a standalone AI Act.[83] Where Australia’s is growing, chiefly by migration, and Treasury projects it will reach some 40.5 million by 2062-2063,[84] AI is not framed as the substitute for the workers but as a form of efficiency.[85] Thus, the extent to which Australia needs to trust AI is only to permit the adoption of AI.
36. Australia does not have any binding risk classification, only guidance as espoused most recently by the National AI Centre’s Guidance for AI Adoption released on October 2025, which condenses the aforementioned ten guardrails into six essential practices: accountability, impact assessment, risk management, information sharing, testing and monitoring, and human oversight.[86]
37. Australia’s only binding AI-specific instrument is for the Commonwealth’s own use rather than for the market.[87]This is seen through the Policy for the Responsible Use of AI in Government requires accountable officials and published transparency statements from federal agencies, and the APS AI Plan of November 2025 extends that architecture across the public service.[88] Yet, these do not reach private deployment, only applying their firmest instrument to the state as purchaser and user, while leaving the private developer to guidance.[89]
38. Australia also has no competent authority, where oversight is dispersed across the Office of the Australian Information Commissioner for privacy, the ACCC for competition and consumer protection, the eSafety Commissioner for online harms, the Australian Human Rights Commission for discrimination, ASIC and the Therapeutic Goods Administration for their sectors, and the Australian Signals Directorate for security.[90] The Australian AI Safety Institute, established with $29.9 million and operational during 2026, tests models, monitors capability and advises — but holds no enforcement power, distinguishing it sharply from a regulator.[91] The Office of AI created in July 2026 is the first Australian attempt at creating a co-ordinating body, yet at the time of writing it has no statutory basis.[92]
39. Lastly (for this article), Australia imposes no AI-specific penalty at all, but the instruments it has chosen to rely on carry maxima several orders of magnitude larger. Serious or repeated interference with privacy attracts the greater of $50 million, three times the benefit obtained, or 30 per cent of adjusted turnover.[93] From 28 March 2026 the fixed limb of the maximum corporate penalty under the Competition and Consumer Act and the Australian Consumer Law doubled to $100 million per contravention.[94] Failure to take reasonable steps under the social media minimum age obligation exposes a platform to $49.5 million.[95] The lesson is that the presence of an AI statute is a poor proxy for regulatory intensity. Thus, while Australia does not have a centralised AI Act, Australia has enacted penalties which heavily regulate the use of AI as well.
IV. Discussion
40. Australia and Singapore initially answered Collingridge by choosing a mix of hard law and soft law, locating which layer of the digital economy each state already binds and controls, and then using soft law to supplement the rest.[96]
41. However, the answer does not appear to be between hard and soft law. Rather, it is any choice which allows the state to act without first having to acquire new authority.[97] This is perhaps due to the multitude of shocks that AI introduces to the society.[98] Where each state decided to legislate based on where societal trust lays, trust is therefore the variable that determines which instruments a society will accept, and the divergence between these jurisdictions is a divergence demonstrates where the allocation of trust in each society.
42. Singapore demonstrates the positive end of this proposition where Singapore has legislated bindingly only where trust was breached.[99] This is a lesson learnt even before the rise of AI, where Parliament restricted police access to TraceTogether data by emergency legislation within a month of the January 2021 disclosure that such access existed, and prohibited digitally manipulated election advertising within months of that harm becoming salient.[100] In a society that treats trust as strategic infrastructure, this was the government’s attempt to repair its citizens’ digital trust.
43. As opposed to this, Singapore’s Model AI Governance Frameworks are neither binding nor do they create any duty on any actor.[101] Thus, compliance becomes a function of the regulatory body’s standing and the trust that actors have in the regulatory body. This demonstrates that organisations follow these frameworks because the public accepts the administrator’s judgment as an adequate substitute for a fixed rule.[102] Thus, the instrument works not because it is better designed than its counterparts elsewhere, but because regulators have the social credit for people to trust in it.[103]
44. Australia demonstrates the converse where Australia has attempted to implement the same structure, yet without the confidence its society has in the regulatory bodies to operate.[104] Australia has implemented voluntary principles and safety standards from 2019 to 2025, with its existing technology-neutral laws and sectoral regulators to remain the foundation in regulating AI.[105] Yet, Australians still recorded the lowest among forty-seven countries surveyed in their willingness to trust AI.[106] The difference thus appears to be the political culture in which these frameworks were introduced into. In a society that litigates readily and expects statutory power rather than advice, advisory and reputational authority carries little weight, especially after the institutional implementation of Robodebt and the repercussions that followed.[107] Australia's chosen institution was therefore the type its political culture respects least, and it could not build the confidence it needed to function.
45. Australia’s change of position validates the proposition, because the change involved a change in the instrument implemented.[108] Before, operative questions including the scope of high-risk activities and adequate practices was left to departmental judgment.[109] The announcement on 15 July 2026 removed the prerogative and discretion from departmental judgment, where the Australian Standards for AI are to be considered by National Cabinet in August 2026 and legislated in early 2027, rather than published as further guidance.[110] A society that does not trust its administration to decide unsupervised will not accept a regime whose content is whatever the administration says it is.[111]Rather, authority such as that belonging to the Parliament must intervene.[112] Indeed, adverse Australia requires binding force especially where administrative judgment is no longer accepted as sufficient.
46. Detractors may argue that congruence with societal expectation is the wrong measure of a legal regime, which presupposes acceptance is equivalent to legitimacy.[113] Against this backdrop, a regime is sound if it matches what its society expects of it.[114] Where the state produces the expectation, the measures implemented thus appear to be the most successful to the government that manages opinions the most effectively rather than one who governs the best.[115] This does not mean that the society has approved of the relevant frameworks. Rather, acquiescence could potentially be recorded as consent.
47. However, this argument assumes that expectations are at the government’s disposal. If this were true, Australia would not need to spend seven years on instruments whose declared purpose was to build public confidence in AI,[116] and yet still be recorded as being the least willing to trust AI. Similarly, while Singapore’s confidence in government stands at 76%,[117] when it emerged in January 2021 that police could access contact-tracing data after ministers had given contrary assurances, this was so egregious that this incident made national news.[118] Parliament restricted access by emergency legislation within a month, and the system was later wound down and the data deleted.[119] These two records are difficult to reconcile with the premise that a government can manufacture the expectations by which it is judged. Every society’s expectations of its government are formed by that government and by the institutions it operates.[120] Instead, analysing the sociology on both state’s contemporary background demonstrates the different constraints operating on both governments, where Australia could not produce the confidence its chosen instrument required, and Singapore could not take the confidence its people had for granted.[121]
48. At least within these jurisdictions, binding law follows the withdrawal of trust rather than the assessment of risk. Trust thus determines the range of instruments a society will accept.
V. Conclusion
49. The intuition that law is the expression of a particular people, not a portable technology, is among the oldest in jurisprudence. Montesquieu insisted in 1748 that the civil and political laws of each nation “should be so appropriate to the people for whom they are made that it is very unlikely that the laws of one nation can suit another.”[122] While Alan Watson showed in recent times that legal rules are in fact transplanted between societies all the time, often easily,[123] Pierre Legrand replied that a rule can never truly travel, because its meaning is inseparable from the culture that gave it sense, so that a “transplant” reduces law to bare propositions stripped of significance.[124]
50. Notably, each jurisdiction takes a deliberate choice as to how they espouse their legal ideals for the society as a whole. Australia’s values are most live in parliament and precedent, checked by litigating regulators and an adversarial civil society.[125] Known as “the only democratic nation in the world without a national charter or bill of rights”,[126] only eight of forty-five proposed constitutional amendments have been approved in the past 125 years.[127] Similarly, Singapore stores its commitments through instruments implemented by the executive which are continuously updated. This includes bills, frameworks, and national strategies.[128] While some instruments appear to bind no one, yet, are still effective where citizens have a high level of trust in the State.[129]
51. Thus, how a society regulates AI should reflect the expectations of the society it governs; where a regime does not yet exist, the choice of whether and how to govern (be it through legislation or guidelines) is itself the reflection.
References
[1] A V Dicey, Lectures on the Relation between Law and Public Opinion in England during the Nineteenth Century (Macmillan 1905) 3.
[2] Constitution of the Republic of Singapore (1985 Rev Ed, 2020 Reprint) art 152(1)-(2).
[3] ibid; art 69 (establishment of the Presidential Council for Minority Rights), art 76 (general function) and art 77 (duty to report to the Speaker whether a Bill contains a differentiating measure — one disadvantageous to a racial or religious community and not equally disadvantageous to others).
[4] On statutory community institutions, Administration of Muslim Law Act 1966 (2020 Rev Ed) (Singapore) ss 75A-75F, establishing the Mosque Building Fund administered by Majlis Ugama Islam Singapura, itself constituted under that Act. On state-sponsored self-help organisations, Central Provident Fund Board (Singapore), ‘Contributions to Self-Help Groups’ <https://www.cpf.gov.sg/employer/employer-obligations/contributions-to-self-help-groups> accessed 26 August 2026: the Board collects contributions from employees’ wages as agent for four community funds — the Chinese Development Assistance Council Fund, the Eurasian Community Fund, the Mosque Building and Mendaki Fund and the Singapore Indian Development Association Fund — administered respectively by CDAC, the Eurasian Association, Muis and SINDA.
[5] Interpretation Act 1965 (2020 Rev Ed) (Singapore) s 9A(1); Tan Cheng Bock v Attorney-General [2017] 2 SLR 850 [37] (Singapore Court of Appeal), setting out the three-step purposive framework, elaborated at [38]-[53] and summarised at [54].
[6] Dicey (n 1) Lecture I, 3-4.
[7] ibid; Lecture I, 3-4 (“Law may be the result of custom”);.
[8] Noam Kolt, ‘Governing AI Agents’ (2026) 101 Notre Dame Law Review (forthcoming) <https://arxiv.org/abs/2501.07913> accessed 15 July 2026.
[9] David Collingridge, The Social Control of Technology (Frances Pinter 1980) 11.
[10] Larry Downes, The Laws of Disruption: Harnessing the New Forces that Govern Life and Business in the Digital Age (Basic Books 2009) 2-3; Gary E Marchant, ‘The Growing Gap Between Emerging Technologies and the Law’ in Gary E Marchant, Braden R Allenby and Joseph R Herkert (eds), The Growing Gap Between Emerging Technologies and Legal-Ethical Oversight: The Pacing Problem (Springer 2011) 19.
[11] Kolt, (n 8).
[12] Kenneth W Abbott and Duncan Snidal, ‘Hard and Soft Law in International Governance’ (2000) 54 International Organization 421.
[13] Gary E Marchant and Brad Allenby, ‘Soft Law: New Tools for Governing Emerging Technologies’ (2017) 73 Bulletin of the Atomic Scientists108, 112.
[14] Ryan Hagemann, Jennifer Huddleston Skees and Adam Thierer, ‘Soft Law for Hard Problems: The Governance of Emerging Technologies in an Uncertain Future’ (2018) 17 Colorado Technology Law Journal 37, 41.
[15] Department of Industry, Science and Resources (Cth), National AI Plan (2 December 2025), confirming reliance on existing technology-neutral laws and sectoral regulators, supported by voluntary guidance and a new AI Safety Institute, in place of a standalone AI Act.
[16] Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence [2024] OJ L1689/1 (“EU AI Act”).
[17] Xinlei Zhao, ‘Technological Hedging and Differentiated Responses of Southeast Asian Countries to U.S.–China Technological Competition: A Case Study on Artificial Intelligence (AI)’ (2025) 38 The Pacific Review 502; Chatham House, How Middle Powers Can Weather US and Chinese AI Dominance (Chatham House Research Paper, February 2026) https://www.chathamhouse.org/2026/02/how-middle-powers-can-weather-us-and-chinese-ai-dominance accessed 15 July 2026.
[18] ibid (n 17) 505-508; EU AI Act (n 16).
[19] ibid (n 17) 502-505.
[20] Eugen Ehrlich, Fundamental Principles of the Sociology of Law (Walter L Moll tr, Harvard University Press 1936), Foreword (“Ehrlich”).
[21] ibid.
[22] Department of Industry, Science and Resources (Cth), National AI Plan (2 December 2025), confirming reliance on existing technology-neutral laws and sectoral regulators, supported by voluntary guidance and a new AI Safety Institute, in place of a standalone AI Act.
[23] ‘Transcript of a Press Conference Given by the Prime Minister of Singapore, Mr Lee Kuan Yew, at Broadcasting House, Singapore, on 9th August 1965’ (National Archives of Singapore) https://www.nas.gov.sg/archivesonline/data/pdfdoc/lky19650809b.pdf accessed 15 July 2026 (transcript recording the pause “for the Prime Minister to regain his composure”) (“Lee”).
[24] ibid. The “little red dot” epithet originates in remarks by President B J Habibie reported in The Asian Wall Street Journal (4 August 1998) and was re-appropriated by Goh Chok Tong in his National Day Rally address of 23 August 1998; on the reappropriation, see Chng Ee Xin, ‘“What’s in a Name?”: The (Un)making of a Singapore Theatre Company’ (2026) 51 Theatre Research International 1, n 3.
[25] Lee (n 23); Shared Values (Cmd 1 of 1991, presented to Parliament 2 January 1991) (Singapore).
[26] Maintenance of Religious Harmony Act 1990 (2020 Rev Ed) (Singapore); the 2019 amendments made restraining orders effective immediately in response to the online spread of falsehoods: Ministry of Home Affairs, ‘First Reading of Maintenance of Religious Harmony (Amendment) Bill’ (2 September 2019) https://www.mha.gov.sg/mediaroom/press-releases/first-reading-of-maintenance-of-religious-harmony-amendment-bill/ accessed 15 July 2026.
[27] Maintenance of Racial Harmony Act 2025 (Singapore), passed 4 February 2025 https://sso.agc.gov.sg/Act/MRHA2025 accessed 15 July 2026.
[28] Shared Values (Cmd 1 of 1991, presented to Parliament 2 January 1991) (Singapore); National Archives of Singapore, ‘White Paper on Shared Values’ https://www.nas.gov.sg/archivesonline/government_records/record-details/a472b486-7aea-11e7-83df-0050568939ad accessed 15 July 2026.
[29] ibid.
[30] On Lee Kuan Yew’s “does it work?” pragmatism, Graham Allison and Robert D Blackwill with Ali Wyne, Lee Kuan Yew: The Grand Master’s Insights on China, the United States, and the World (MIT Press 2013); critically, Kenneth Paul Tan, ‘The Ideology of Pragmatism: Neo-liberal Globalisation and Political Authoritarianism in Singapore’ (2012) 42 Journal of Contemporary Asia 67.
[31] Ministry of Defence (Singapore), ‘Total Defence’ <https://www.mindef.gov.sg/defence-matters/defence-topics/total-defence/> accessed 15 July 2026.
[32] ibid.
[33] Ministry of Defence (Singapore), ‘Total Defence’ https://www.mindef.gov.sg/defence-matters/defence-topics/total-defence/ accessed 15 July 2026 (five pillars from 1984); Ministry of Defence (Singapore), ‘Total Defence Strengthened with Addition of Digital Defence as the Sixth Pillar’ (15 February 2019) https://www.mindef.gov.sg/news-and-events/latest-releases/15feb19_nr accessed 15 July 2026.
[34] ibid.
[35] Ministry of Home Affairs (Singapore), ‘Oral Reply to Parliamentary Question on Whether TraceTogether Data Will Be Used for Criminal Investigations’ (4 January 2021) <https://www.mha.gov.sg/media-room/newsroom/oral-reply-to-parliamentary-question-on-whether-tracetogether-data-will-be-used-for-criminal-investigations-by-mr-desmond-tan-minister-of-state-ministry-of-home-affairs-and-ministry-of-sustainability-and-the-environment/> accessed 26 August 2026 (reply of Desmond Tan, Minister of State for Home Affairs).
[36] COVID-19 (Temporary Measures) (Amendment) Act 2021 (Act 3 of 2021) (Singapore), introduced under a certificate of urgency on 1 February 2021 and passed on 2 February 2021, inserting Pt 11 and the Seventh Schedule (seven categories of serious offence) and imposing a deletion obligation on the data administrator; Ministry of Digital Development and Information, ‘Amendments in COVID-19 (Temporary Measures) Act on the Use of Personal Digital Contact Tracing Data’ (1 February 2021) <https://www.mddi.gov.sg/newsroom/use-of-personal-digital-contact-tracing-data/> accessed 20 August 2026.
[37] Edelman, 2026 Edelman Trust Barometer: Singapore Report (March 2026) 38 (trust in government 76%, down one point year on year; my employer 70%, business, media and NGOs 60% each).
[38] ‘Singapore’s Smart Leap: Digital Minister Teo on AI Transformation’ (McKinsey & Company, 5 January 2026) <https://www.mckinsey.com/featured-insights/future-of-asia/singapores-smart-leap-digital-minister-teo-on-ai-transformation> accessed 12 July 2026 (“Teo”).
[39] ibid.
[40] Smart Nation and Digital Government Office (Singapore), National Artificial Intelligence Strategy 2.0: AI for the Public Good, for Singapore and the World (December 2023) (“National AI Strategy 2.0”).
[41] Teo (n 38).
[42] ibid.
[43] National AI Strategy 2.0 (n 40).
[44] ibid; Teo (n 38).
[45] Personal Data Protection Commission, Model Artificial Intelligence Governance Framework (2nd edn, 21 January 2020) 4, 20 <https://www.pdpc.gov.sg/-/media/files/pdpc/pdf-files/resource-for-organisation/ai/sgmodelaigovframework2.pdf> accessed 12 July 2026 (“Model AI Governance Framework”).
[46] ibid 4, 20.
[47] Teo (n 38).
[48] Model AI Governance Framework for Agentic AI (n 45).
[49] ibid; 4, 20.
[50] AI Verify Foundation and Infocomm Media Development Authority, Model AI Governance Framework for Generative AI (30 May 2024) <https://aiverifyfoundation.sg/wp-content/uploads/2024/05/Model-AI-Governance-Framework-for-Generative-AI-May-2024-1-1.pdf> accessed 12 July 2026.
[51] Ministry of Digital Development and Information, ‘Singapore Launches New Model AI Governance Framework for Agentic AI’ (22 January 2026) <https://www.mddi.gov.sg/newsroom/singapore-launches-new-model-ai-governance-framework-for-agentic-ai–/> accessed 12 July 2026; Infocomm Media Development Authority, Model AI Governance Framework for Agentic AI (ver 1.5, 20 May 2026) <https://www.imda.gov.sg/-/media/imda/files/about/emerging-tech-and-research/artificial-intelligence/mgf-for-agentic-ai.pdf> accessed 12 July 2026 (“Model AI Governance Framework for Agentic AI”)..
[52] Model AI Governance Framework for Agentic AI (n 51); Model AI Governance Framework for Generative AI (n 45); National AI Strategy 2.0 (n 40).
[53] Ministry of Digital Development and Information, ‘Singapore Launches World’s First AI Testing Framework and Toolkit to Promote Transparency’ (25 May 2022) <https://www.mddi.gov.sg/newsroom/sg-launches-first-ai-testing-framework-and-toolkit/> accessed 12 July 2026.
[54] ibid.
[55] Infocomm Media Development Authority, ‘Singapore Launches AI Verify Foundation to Shape the Future of International AI Standardsthrough Collaboration’ (7 June 2023) <https://www.imda.gov.sg/resources/press-releases-factsheets-and-speeches/press-releases/2023/singapore-launches-ai-verify-foundation> accessed 12 July 2026.
[56] National Institute of Standards and Technology, ‘Crosswalk Documents’ (NIST AI Resource Center) <https://airc.nist.gov/airmf-resources/crosswalks/> accessed 12 July 2026; Josh Lee Kok Thong, ‘Explaining the Crosswalk Between Singapore’s AI Verify Testing Framework and the US NIST AI Risk Management Framework’ (Future of Privacy Forum, 23 January 2024) <https://fpf.org/blog/explaining-the-crosswalk-between-singapores-ai-verify-testing-framework-and-the-u-s-nist-ai-risk-management-framework/> accessed 12 July 2026.
[57] Ministry of Finance (Singapore), Budget 2026 Statement (12 February 2026), announcing a National AI Council to be chaired by the Prime Minister; ‘Budget 2026: PM Wong to Chair New National AI Council to Drive Transformation in Key Sectors’ The Straits Times (Singapore, 12 February 2026).
[58] ibid.
[59] ibid.
[60] ibid.
[61] Ministry of Health, Health Sciences Authority and Integrated Health Information Systems, Artificial Intelligence in Healthcare Guidelines (AIHGle) (October 2021), updated as AIHGle 2.0 (13 March 2026) <https://www.hcsa.gov.sg/licensable-healthcare-services/moh-mhc-cir-0024-2026-updated-moh-artificial-intelligence-in-healthcare-guidelines–aihgle-2-0-/> accessed 12 July 2026.
[62] Personal Data Protection Commission, Advisory Guidelines on Use of Personal Data in AI Recommendation and Decision Systems (1 March 2024) <https://www.pdpc.gov.sg/organisations/regulations-decisions/regulatory-guidance/advisory-guidelines-on-use-of-personal-data-in-ai-recommendation-and-decision-systems> accessed 12 July 2026.
[63] Elections (Integrity of Online Advertising) (Amendment) Act 2024 (No 34 of 2024) (Singapore) <https://sso.agc.gov.sg/Acts-Supp/34-2024/Published/20241121?DocDate=20241121> accessed 12 July 2026.
[64] Elections (Integrity of Online Advertising) (Amendment) Act 2024 (No 34 of 2024) (Singapore) <https://sso.agc.gov.sg/Acts-Supp/34-2024/Published/20241121?DocDate=20241121> accessed 12 July 2026.
[65] ASEAN, ASEAN Guide on AI Governance and Ethics (ASEAN 2024) <https://asean.org/book/asean-guide-on-ai-governance-and-ethics/> accessed 12 July 2026.
[66]ibid; ASEAN, Expanded ASEAN Guide on AI Governance and Ethics – Generative AI (ASEAN 2025) <https://asean.org/book/expanded-asean-guide-on-ai-governance-and-ethics-generative-ai/> accessed 12 July 2026; Ministry of Digital Development and Information, ‘Singapore Concludes Fruitful Chairmanship of the ASEAN Digital Ministers’ Meeting’ <https://mddi.gov.sg/media-centre/press-releases/singapore-concludes-fruitful-chairmanship-of-adgmin> accessed 12 July 2026.
[67] Australian Bureau of Statistics, ‘Cultural Diversity: Census, 2021’ (2022) https://www.abs.gov.au/statistics/people/people-and-communities/cultural-diversity-census/2021 accessed 7 August 2026 (51.5% of Australians born overseas or with a parent born overseas); Australian Bureau of Statistics, ‘2021 Census Shows Changes in Australia’s Religious Diversity’ (media release, 28 June 2022) https://www.abs.gov.au/media-centre/media-releases/2021-census-shows-changes-australias-religious-diversity accessed 7 August 2026 (no religion 38.9%; Christianity 43.9%, down from 61.1% a decade earlier).
[68] Herbert Smith Freehills Kramer, ‘Class Actions Radar: Australia’ <https://www.hsfkramer.com/insights/reports/class-actions-radar-scanning-global-trends-and-risks/australia> accessed 7 August 2026.
[69] ibid; Joe Longo, ‘We’re Not There Yet: Current Regulation around AI May Not Be Sufficient’ (ASIC, speech, 31 January 2024) https://www.asic.gov.au/about-asic/news-centre/speeches/we-re-not-there-yet-current-regulation-around-ai-may-not-be-sufficient/ accessed 7 August 2026.
[70] ibid.
[71] Nicole Gillespie, Steve Lockey, Tiarne Ward, Alayna Macdade and Gerard Hassed, Trust, Attitudes and Use of Artificial Intelligence: A Global Study 2025 (University of Melbourne and KPMG 2025) DOI 10.26188/28822919; University of Melbourne and KPMG, Trust, Attitudes and Use of Artificial Intelligence: A Global Study 2025 — Australia Insights (2025) <https://assets.kpmg.com/content/dam/kpmg/au/pdf/2025/trust-in-ai-global-insights-2025-australia-snapshot.pdf> accessed 26 August 2026.
[72] Prygodicz v Commonwealth of Australia (No 2) [2021] FCA 634 [1]-[9] (Murphy J), approving settlement of the class action; Royal Commission into the Robodebt Scheme, Report (July 2023).
[73] Royal Commission (n 72) 373, finding that the scheme undermined trust in government and confidence in the social security system; Attorney-General (Cth), ‘Robodebt Class Action Settlement’ (Media Release, 4 September 2025) (a further $475 million in compensation, described as the largest class action settlement in Australian history).
[74] Herbert Smith Freehills Kramer (n 68); Prygodicz (n 72).
[75] National AI Plan (n 15).
[76] See eg Privacy Act 1988 (Cth); Competition and Consumer Act 2010 (Cth); Online Safety Act 2021 (Cth).
[77] Department of Industry, Innovation and Science (Cth), Australia’s Artificial Intelligence Ethics Framework: AI Ethics Principles (November 2019), setting out eight voluntary principles.
[78] Department of Industry, Science and Resources (Cth), Safe and Responsible AI in Australia: Proposals Paper for Introducing Mandatory Guardrails for AI in High-Risk Settings (September 2024), proposing ten mandatory guardrails and canvassing three implementation options — sector-specific amendment, framework legislation, or a whole-of-economy AI Act with a dedicated regulator.
[79] National AI Plan (n 15).
[80]Anthony Albanese, ‘AI in Australia’s Interests’ (Speech, University of Sydney, 15 July 2026); Prime Minister of Australia, ‘AI in Australia’s Interests’ (Media Release, 15 July 2026) <https://www.pm.gov.au/media/ai-australias-interests> accessed 20 August 2026; Department of the Prime Minister and Cabinet (Cth), ‘Office of AI’ <https://www.pmc.gov.au/domestic-policy/office-ai> accessed 20 August 2026 (Office established 15 July 2026; National Cabinet to consider the framework in August 2026; standards expected to be legislated in early 2027).
[81] Department of the Prime Minister and Cabinet (n 80) (mandatory requirements for large AI data centres, including energy and water standards, with copyright protections for Australian creators).
[82] Department of the Prime Minister and Cabinet (n 80): the standards are to impose mandatory requirements for large AI data centres, including energy and water standards, together with copyright protections for Australian creators. Legislation is expected in early 2027, when the final scope will be settled
[83] National AI Act (n 15).
[84]Commonwealth Treasury, Intergenerational Report 2023: Australia’s Future to 2063 (August 2023) 4–12; Australian Bureau of Statistics, ‘Australia’s Population Grows by 2.5%’ (21 March 2024) (net overseas migration accounting for the large majority of annual growth).
[85] National AI Plan (n 15).
[86] National Artificial Intelligence Centre (Cth), Guidance for AI Adoption (21 October 2025), issued in ‘Foundations’ and ‘Implementation Practices’ versions and condensing the earlier ten guardrails into six essential practices.
[87] Digital Transformation Agency (Cth), Policy for the Responsible Use of AI in Government (September 2024, updated December 2025), requiring accountable officials and published transparency statements.
[88] ibid; Australian Public Service Commission, APS AI Plan (November 2025).
[89] ibid.
[90]Privacy Act 1988 (Cth); Competition and Consumer Act 2010 (Cth); Online Safety Act 2021 (Cth); Corporations Act 2001 (Cth) and ASIC Regulatory Guide 255; Therapeutic Goods (Medical Devices) Regulations 2002 (Cth).
[91]Minister for Industry and Innovation, ‘Establishment of Australian AI Safety Institute’ (Media Release, November 2025); the Institute became operational in 2026 with $29.9 million in committed funding and carries testing, monitoring and advisory functions but no enforcement power.
[92] Department of the Prime Minister and Cabinet (n 80); enabling legislation is expected in early 2027.
[93]Privacy Act 1988 (Cth) s 13G, as amended by the Privacy Legislation Amendment (Enforcement and Other Measures) Act 2022 (Cth) (the greater of $50 million, three times the benefit obtained, or 30 per cent of adjusted turnover in the relevant period).
[94]Treasury Laws Amendment (Doubling Penalties for ACCC Enforcement) Act 2026 (Cth) No 19 of 2026, commenced 28 March 2026, raising the first limb of the maximum corporate penalty under the Competition and Consumer Act 2010 (Cth) and the Australian Consumer Law from $50 million to $100 million per contravention.
[95]Online Safety Act 2021 (Cth) pt 4A, with civil penalties of up to $49.5 million for failure to take reasonable steps under the social media minimum age obligation.
[96] Collingridge (n 9) 11.
[97] Abbott and Snidal (n 9) 421-422.
[98] Kolt (n 8).
[99] COVID-19 (Temporary Measures) (Amendment) Act 2021 (n 36); Elections (Integrity of Online Advertising) (Amendment) Act 2024 (n 63).
[100] ibid.
[101] Model AI Governance Framework (n 45); Model AI Governance Framework for Generative AI (n 45); Model AI Governance Framework for Agentic AI (n 51).
[102] Teo (n 38).
[103] Edelman (n 37).
[104] AI Ethics Principles (n 77); Guidance for AI Adoption (n 86).
[105] ibid; National AI Plan (n 15).
[106] Gillespie and others (n 71) 32, 45.
[107] Herbert Smith Freehills Kramer (n 68); Prygodicz (n 72).
[108] Albanese (n 80).
[109] Guidance for AI Adoption (n 86).
[110] Prime Minister of Australia, ‘AI in Australia’s Interests’ (Media Release, 15 July 2026) <https://www.pm.gov.au/media/ai-australias-interests> accessed 20 August 2026; Department of the Prime Minister and Cabinet (Cth), ‘Office of AI’ <https://www.pmc.gov.au/domestic-policy/office-ai> accessed 20 August 2026 (Office established 15 July 2026; National Cabinet to consider the approach in August 2026; standards expected to be legislated in early 2027).
[111] Australia Insights (n 71) 1 (77% consider AI regulation to be required; 30% consider current safeguards sufficient).
[112] ibid.
[113] Lon L Fuller, The Morality of Law (rev edn, Yale University Press 1969) ch 2 (“The Morality that Makes Law Possible”), where congruence between official action and declared rule is one of the eight conditions of legality — congruence being a criterion internal to law rather than a measure of popular acceptance.
[114] ibid.
[115] ibid.
[116] Department of Industry, Innovation and Science (Cth), Australia’s Artificial Intelligence Ethics Framework: AI Ethics Principles (November 2019); Department of Industry, Science and Resources (Cth), Safe and Responsible AI in Australia: Proposals Paper for Introducing Mandatory Guardrails for AI in High-Risk Settings (September 2024); National Artificial Intelligence Centre (Cth), Guidance for AI Adoption (21 October 2025); Department of Industry, Science and Resources (Cth), National AI Plan (2 December 2025), whose third stated goal is keeping Australians safe.; Minister for Industry and Innovation, ‘Establishment of Australian AI Safety Institute’ (Media Release, November 2025); the Institute became operational during 2026 with $29.9 million in committed funding and carries testing, monitoring and advisory functions but no enforcement power.
[117] Edelman, 2026 Edelman Trust Barometer: Singapore Report (March 2026) (trust in government 76%).
[118] https://www.channelnewsasia.com/singapore/tracetogether-data-use-police-restricted-very-serious-offences-says-shanmugam-5658946; https://www.mha.gov.sg/media-room/newsroom/oral-reply-to-parliamentary-question-on-whether-tracetogether-data-will-be-used-for-criminal-investigations-by-mr-desmond-tan-minister-of-state-ministry-of-home-affairs-and-ministry-of-sustainability-and-the-environment/; https://ink.library.smu.edu.sg/cgi/viewcontent.cgi?article=5641&context=sol_research
[119] COVID-19 (Temporary Measures) (Amendment) Act 2021 (Singapore), passed within a month of the January 2021 disclosure that police could access TraceTogether data, restricting access to seven categories of serious offence; the system was subsequently wound down and the data deleted in 2023.
[120] Ehrlich (n 7) Foreword; Legrand (n 24).
[121] Gillespie and others (n 71); Edelman (n 537); COVID-19 (Temporary Measures) (Amendment) Act 2021 (n 36).
[122] Montesquieu, The Spirit of the Laws (Anne M Cohler, Basia Carolyn Miller and Harold Samuel Stone eds and trs, CUP 1989) bk 1 ch 3, 8.
[123] Alan Watson, Legal Transplants: An Approach to Comparative Law (2nd edn, University of Georgia Press 1993).
[124] Pierre Legrand, ‘The Impossibility of “Legal Transplants”’ (1997) 4 Maastricht Journal of European and Comparative Law 111.
[125] Herbert Smith Freehills Kramer (n 68); Longo (n 69).
[126] George Williams and Daniel Reynolds, A Charter of Rights for Australia (4th edn, NewSouth 2017); ‘A Charter of Rights for Australia “More Urgent than Ever”’ (UNSW Newsroom, 2 August 2017) https://www.unsw.edu.au/newsroom/news/2017/08/a-charter-of-rights-for-australia--more-urgent-than-ever-- accessed 7 August 2026. Only the ACT, Victoria and Queensland maintain statutory charters.
[127] Australian Electoral Commission, ‘Referendum Dates and Results’ https://www.aec.gov.au/elections/referendums/ accessed 7 August 2026; the fourteenth of October 2023 Voice referendum was defeated 60.06% to 39.94% nationally: AEC, ‘2023 Referendum National Results’ https://results.aec.gov.au/29581/Website/ReferendumNationalResults-29581.htm accessed 7 August 2026.
[128] National AI Strategy 2.0 (n 43); Model AI Governance Framework for Agentic AI (n 48); Model AI Governance Framework (n 51); Model AI Governance Framework for Generative AI (n 66); Maintenance of Racial Harmony Act 2025 (n 26).
[129] Edelman (n 37); Teo (n 38).
Comments