Paterson, Jeannie --- "Misleading or Deceptive AI: Why Transparency Provides the Baseline for Responsible AI in Consumer Transactions" [2024] UNSWLawSocCConsc 16; (2024) 18 UNSW Law Society Court of Conscience 113

---

MISLEADING OR DECEPTIVE AI:

WHY TRANSPARENCY PROVIDES THE BASELINE FOR RESPONSIBLE AI IN CONSUMER TRANSACTIONS

Jeannie Paterson^*

I Introduction

Artificial intelligence (‘AI’) has been growing in prominence in modern society, especially with the emergence of generative AI, such as in ChatGPT. While AI offers many benefits, it also carries significant risks of harm. These risks can be classified in several ways but include potential harms relating to discrimination, privacy, misinformation, safety, human-computer interactions and social/environmental impact.^[1] In response to concerns about these risks, there have been a growing number of initiatives worldwide aimed at ensuring robust and accountable human governance of AI. These initiatives cover ethical principles,^[2] technical standards,^[3] and new laws directed specifically at AI,^[4] the most prominent example being the European Union’s (‘EU’) AI Act.^[5] In Australia, the Commonwealth Government has put forward a number of initiatives to support responsible AI, most recently, a ‘Voluntary AI Safety Standard’^[6] and a law reform proposal for ‘Mandatory Guardrails for Responsible AI’.^[7]

There is insufficient space in this short article to discuss all of these issues or initiatives. Instead, this article focuses on one common element of most frameworks for responsible AI, transparency. It focuses on the role of transparency in business-to-consumer transactions as a specific high-impact site of operation of AI, noting these insights may prove relevant to other contexts in which AI is deployed. The article seeks to explain why transparency is a necessary requirement for AI accountability, albeit not sufficient, in transactions involving consumers, and the way in which transparency requirements complement existing laws for consumer protection. It illustrates these arguments by considering three ways in which AI may interact with consumers, namely AI used by consumers to inform their decisions; AI used to make decisions about consumers’ access to goods and services; and AI used against consumers to manipulate or trick them. The article begins by outlining initiatives for transparency in AI models and systems, before turning to these specific scenarios.

II Understanding Transparency As A Prerequisite To Responsible Ai

Principles, standards and laws for responsible AI commonly require transparency across the lifecycle of that technology. Transparency underlines many of the reporting and documentation requirements in the EU AI Act.^[8] Thus, Australia’s AI Ethics Principles state that there ‘should be transparency and responsible disclosure so people can understand when they are being significantly impacted by AI, and can find out when an AI system is engaging with them’.^[9] The Australian Voluntary AI Safety Standards includes both informing ‘end-users regarding AI-enabled decisions, interactions with AI and AI-generated content’ and being ‘transparent with other organisations across the AI supply chain about data, models and systems to help them effectively address risks’.^[10] Australia’s Proposed Mandatory Guardrails for Responsible AI include measures for ‘transparency regarding product development and use with end-users, other actors in the AI supply chain and relevant authorities’.^[11]

As these principles demonstrate, transparency in the context of responsible AI may operate at a number of levels of generality.^[12] Recital 27 of the EU AI Act explains:

transparency means that AI systems are developed and used in a way that allows appropriate traceability and explainability, while making humans aware that they communicate or interact with an AI system, as well as duly informing deployers of the capabilities and limitations of that AI system and affected persons about their rights.^[13]

Thus, ‘transparency’ as an AI governance principle can refer to clarity about the data and algorithms used in AI systems. This kind of transparency is a necessary precondition to identifying problems of bias or error in AI outputs. This is because these failings in an AI system will often only be able to be identified by insight into the way in which AI was developed and trained, the data used in training and the outputs it produces. It is these insights that allow the testing, monitoring or auditing needed to identify patterns of discrimination or error.^[14] For the same reasons, transparency at this level is an essential part of holding accountable those who provide or deploy AI for safe, secure and effective systems.

As the EU AI Act Recital recognises, transparency also operates at a more general level to encompass informing humans when they are interacting with AI and the basis for decisions made about them by, or informed by, AI.^[15] In the context of business to consumer transactions, AI transparency means consumers should know when they are dealing with, subject to or interacting with content generated by AI. Informing humans about the way AI is being used in interactions with them respects human dignity and autonomy, in this instance manifested as a right for consumers not to be misled about who or what they are interacting with, as well as to exercise choice about the kinds of interactions they prefer to engage in. Additionally, transparency about the interactions between consumers and AI is essential for regulators and consumer advocates to oversee the operation of AI systems and their impact on consumers’ rights and interests.^[16]

New initiatives for transparency in AI overlap with existing requirements in Australian law. In particular, the Competition and Consumer Act 2010 (Cth) sch 2 (‘Australian Consumer Law’) contains broad prohibitions on misleading conduct in trade or commerce.^[17] Misleading conduct contrary to this statutory prohibition can be found in failing to reveal information that consumers would reasonably expect to be disclosed.^[18] This aspect of the statutory prohibition may overlap with transparency requirements. In particular, firms that fail to disclose the use of AI in circumstances where consumers might reasonably think they are dealing with a human may be engaged in misleading conduct contrary to the statutory regime.

There is nonetheless a limitation on using prohibitions on misleading conduct to prompt transparency in AI used in business-to-consumer transactions. This limitation arises because the prohibition is a response to harmful conduct rather than a mechanism for AI accountability, which aims to prevent harm from arising in the first place. A liability regime may provide incentives to firms to take more care in their operations, including in their use of AI, from the outset. Indeed deterrence (against contravening conduct) is one of the main purposes of the courts’ powers to impose civil pecuniary penalties for some contraventions of the Australian Consumer Law.^[19] However, given the complexity of AI, and the speed with which it is developing, we may not want to wait until harm occurs to see if the case law encourages firms to be more transparent in their use of AI. This need for transparency and other protections, applying ex ante, as a condition of use and before harm occurs, is one of the reasons for considering mandatory rules applying to AI in Australia.^[20]

III Transparency In Consumer Ai Transactions

To understand the case for requiring transparency, ex-ante, in business-to-consumer transactions, it is useful to consider three key categories of AI use which may impact on consumer interests and the role of transparency responding to the risks of harm arising from those uses.

A AI Used by Consumers

The first relevant category of use involves AI used by consumers to provide advice and assistance, such as through apps or smart speakers. The AI in this scenario may be embedded in an object (a speaker, phone or household item) but is typically provided as a service.^[21] The Australian Consumer Law provides a regime of consumer guarantees which are mandatory standards of quality that apply to the supply of goods or services to consumers.^[22] Under s 60 of this regime, a provider of services (in this case the firm deploying the AI to provide the service to consumers) is required to use ‘due care and skill’ in providing that service. What amounts to ‘due care’ for these purposes is likely to be guided by the governance and accountability strategies set out in guidance such as the Australian Voluntary AI Safety Standard.^[23] This means firms providing AI services to consumers should have systems and processes for reducing the possible risk of harm to consumers that may arise specifically from the use of AI, including by enhancing privacy and security, and responding to possible biased or self-serving recommendations along with systemic error.^[24] Transparency lies in documenting what has been done to satisfy these requirements. A commitment to transparency may sometimes also go further to require salient, clear and accessible advice to consumers about the limitations of the technology, such as, in services based on generative AI, the tendency to hallucinate.

Another aspect of the concern with AI used by consumers arises in respect to AI bots and interfaces designed to resemble humans, a scenario described by Frank Pasquale as counterfeiting humans.^[25] Here, the risk is that consumers may be mistaken about the status of a humanistic interface, thinking they are interacting with another human.^[26] It will usually be wrong, in law and normatively, to mislead humans about the sentience of the entity they are dealing with.^[27] It is also possible to envisage a human being manipulated as a result of this counterfeited relationship.^[28] The human may put misplaced trust in the AI to act in their best interests, when the AI has no capacity to care or even do what is best for the human simply because it is a mere machine.^[29]

This concern for transparency in AI that might otherwise be mistaken for a human is recognised in some legislation. The California Bolstering Online Transparency Act requires that bots used to influence a vote or incentivise a transaction be expressly identified as mere bots.^[30] The EU AI Act imposes transparency obligations on providers of AI systems to ensure humans ‘are informed that they are interacting with an AI system, unless this is obvious’.^[31] The voluntary Australian AI Safety Standard recommends deployers of AI systems consider transparency to mitigate risks associated with AI systems ‘presenting as a person’.^[32] At the time of writing, the question in Australian law reform is whether these requirements are sufficiently important to be embodied in law (i.e. made mandatory).

B AI Used to Make Decisions about Consumers

The second category of AI use requiring transparency in business-to-consumer transactions arises when AI is used to make or inform decisions about consumers, commonly referred to as ‘automated’ or ‘algorithmic’ decision-making. This use of AI involves using the predictive insights of machine learning algorithms drawn from data sets to personalise aspects of the price or supply of goods and services to consumers. For example, AI might be used to set the interest rate for a loan, the cost of insurance or access to education or healthcare programs. These predications might benefit consumers by ensuring lower prices or more suitable services. However, there are also concerns that relying on algorithms to inform such matters may be used in a predatory manner against inexperienced or vulnerable consumers to make them pay more or exclude them from the market altogether. Additionally, these outcomes may be based on factors that might be considered irrelevant or at least not compelling in the context, such as social media ‘likes’ or common items in a shopping basket.

The risks of harm inherent in automated decision-making, and the need for governance measures to reduce the risk of those harms, were starkly revealed in Australia by the robo-debt scandal.^[33] Decisions informed by AI might be more precise and accurate than if made by humans, but they may also lack the safeguard that is provided by human insight and oversight. Where training data gives rise to errors or discriminatory biases, the harmful effects of these influences can be magnified by automation and difficult to identify because they are embedded in an opaque system.^[34]

In most cases firms are not required to provide goods or services to consumers, and the prices at which those products should be supplied are not fixed by law. This means that there are very few methods in consumer protection law for scrutinising the decision not to supply a product to consumers, or the price of a product that is supplied. Of course, if a particular representation has been made about the availability of the product or price, a firm will be held to it.^[35] Generally, however, matters of pricing or supply are typically ‘back of house’ decisions to which the protections in the Australian Consumer Law are unlikely to apply because they do not involve ‘supply’ in ‘trade or commerce’.^[36] Yet we as a society might think that firms using AI for these kinds of decisions about consumers should be accountable for those decisions, at least insofar as ensuring they are not perpetuating unacceptable biases or being influenced by irrelevant or unacceptable factors.^[37]

A first step towards this accountability might lie in requiring firms to be transparent about their use of automated decision making, and the factors relevant to the automated decision. Reforms to the Privacy Act 1988, announced in September 2024, go in this direction. The Privacy and Other Legislation Amendment Bill 2024 (Cth), being considered by parliament at the time of writing, proposes amendments to the requirement to provide notice under Australian Privacy Principle 1. These proposed reforms provide that, where an entity subject to the Act uses a ‘computer program’ to ‘make, or do a thing that is substantially and directly related to making, a decision’ that could 'reasonably be expected to significantly affect the rights or interests’ of an individual, they must include in the privacy notice details about the automated process and the personal information that is ‘directly or substantially’ related to the decision.^[38] These proposed transparency requirements do not include a right to not to be subject to automated decision-making.^[39] However, they may provide the foundation for other kinds of challenge to unfair decisions, and, ideally, would complement other reforms, such as requirements for accountability and fairness in the voluntary/mandatory guardrails for responsible AI in Australia.^[40]

C AI Used Against Consumers

Deepfakes are images, videos or audio that falsely present real people saying or doing things they did not do.^[41] They may do this by manipulating aspects of the image or voice of the person, or, more recently, by generating an entirely synthetic image, video or audio.^[42] Most concern has been raised about deepfake images, which are only magnified by the addition of seemingly realistic voice cloning.^[43] Deepfakes can be entirely legitimate, such as in film and television. Yet deepfakes also have the potential to cause considerable individual, economic and democratic harm.^[44] Deepfakes have been implicated in intimate image abuse,^[45] scams^[46] and political misinformation.^[47] They risk an erosion of trust in public institutions. Chesney and Citron also point to the ‘liars’ dividend’, whereby greater concern about deepfakes creates distrust in all images and leads to a scepticism about genuine images.^[48]

The malicious use of deep fake content may contravene existing laws, such as, potentially, defamation, deceit and misrepresentation. The nonconsensual creation and sharing of intimate images, including deep fake or synthetic images, is a criminal offence in most Australian jurisdictions, with new federal offences recently introduced.^[49] Victims of offensive deepfake images can request that platforms and websites remove the images. In Australia, the Online Safety Act 2021 (Cth) provides the eSafety Commissioner with a power to require that offending images within the scope of that act (primarily child abuse material, non-consensual intimate images, abhorrent violence and terrorist material) be removed from websites and platforms, accompanied by fines for those who do not comply.^[50]

Unfortunately, the wrongdoer in cases of deepfake fraud or scams will usually have disappeared. This reality means that a regulatory response to the harms of deep fakes is likely to need to focus on the firms involved in the creation and transmission of deepfakes, such as developers of the tools that create deep fakes, app stores that distribute the tools and platforms that display the images. The regulatory challenge is to provide realistic incentives to these firms to proactively respond to the risks arising from deepfake creation and sharing.^[51] The Australian Government has recently announced a suite of reforms aimed at scams (including through deepfakes)^[52] and misinformation.^[53] In relation to the consumer protection issue of scams, the proposed reforms include placing greater obligations on intermediaries such as digital platforms, banks and telecommunications companies for preventing, detecting and disrupting scams, with significant civil penalties for failing to comply with these obligations, as well as, in the case of digital platforms, wider takedown obligations.^[54]

Importantly, these responses require mechanisms for identifying fake content. Some regulatory regimes are imposing obligations for greater transparency in the provenance of online content. Thus, under the EU AI Act, Article 50 states that:

2. Providers of AI systems, including general-purpose AI systems, generating synthetic audio, image, video or text content, shall ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated ...

4. Deployers of an AI system that generates or manipulates image, audio or video content constituting a deep fake, shall disclose that the content has been artificially generated or manipulated.

In the US, the White House has obtained a voluntary commitment from tech companies to develop ‘robust technical mechanisms to ensure users know when content is AI-generated, such as a watermarking system’.^[55]

These developments are important in requiring providers of AI systems to use technological methods to allow deepfakes to be identified. These mechanisms include ‘trust marks’ and ‘watermarks’. In particular, the Coalition for Content Providence and Authenticity (‘CCPA’) has developed a technical standard for ‘content provenance and authenticity’.^[56] This technical standard includes a ‘manifest’ that identifies the source of digital images using cryptographically signed metadata and that can be used to verify the provenance of images and subsequent changes to them.^[57] The CCPA has also developed standards for watermarking.^[58] Watermarks create a more ‘sticky’ link between the digital content and its origins by being embedded in the synthetic image, video or audio track which is then detectable by software.^[59] In the case of the CCPA watermark, this links back to the providence metadata credentials included in the trust mark or manifest, even if this information has been subsequently removed from the image. Additionally, there are some initiatives to preserve the providence of authentic images, even storing the provenance data on a blockchain.^[60]

These strategies are nascent, and there is considerable debate in the tech community about their efficacy.^[61] Trust and water marks can be manipulated by bad actors. They require considerable coordination across the digital ecosystem to be recognised and used. The overall strategy for monitoring the provenance of images need to be carefully designed and monitored so as not to censor unpopular opinions or the views of marginalised groups, or stifle the use of satire and humour in public debate.^[62] Nonetheless, they represent one part of a suite of possible responses to the harms of AI deep fakes.^[63] They also provide a strategy that potentially can inform and empower consumers rather than leaving them at the mercy of fake images. The reality is that these issues of AI used against consumers can only begin to be addressed by an ongoing coordinated effort between regulators, firms, government and consumers. The technology, and its possible misuses, develop more quickly than rule-based law reform can ever accommodate.

IV Conclusion

Greater transparency is not a fail-safe response to the problems of misleading or deceptive AI in consumer transactions. It will not prevent consumers from being misled about the nature of the entity they deal with online. Not will it preclude firms or agencies using AI to inform decisions based on factors that may be considered irrelevant or leading to biased outcomes. Transparency will not prevent deepfakes from being used for scams, fraud and abuse. However, transparency in these domains is the prerequisite for making effective more sweeping obligations on those who develop, provide or deploy AI, including through testing, training and oversight. Law reform to introduce proactive and ex ante transparency requirements for AI used by, about and against consumers is worth supporting even if it needs to be buttressed by other reforms and initiatives.

---

^* Jeannie Paterson is a Professor of Law and Director of the Centre of AI and Digital Ethics at the University of Melbourne.

¹ See Peter Slattery et al, The AI Risk Repository: A Comprehensive Meta-Review, Database, and Taxonomy of Risks From Artificial Intelligence (Report, 2024). See also Digital Platforms Regulators Forum, ‘Literature Summary: Harms and Risks of Algorithms’ (Working Paper No 1, Digital Platforms Regulators Forum, July 2023).

^[2] ‘Australia’s AI Ethics Principles’, Department of Industry, Science and Resources (Web Page, 2019) <https://www.industry.gov.au/publications/australias-artificial-intelligence-ethics-framework/australias-ai-ethics-principles> (‘Australia’s AI Ethics Principles’).

^[3] See Standards Australia, ‘Standards Australia Adopts the International Standard for AI Management System, AS ISO/IEC 42001:2023’ (Media Release, 16 February 2024) <www.standards.org.au/news/standards-australia-adopts-the-international-standard-for-ai-management-system-as-iso-iec-42001-2023>.

^[4] See the summary of legal initiatives in 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 (Proposals Paper, September 2024) 3 (‘Proposed Mandatory Guardrails’) .

^[5] Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 Laying Down the Harmonised Rules of Artificial Intelligence and Amending Regulation [2024] OJ L 2024/1689 (‘EU AI Act’).

^[6] Department of Industry, Science and Resources (Cth), Voluntary AI Safety Standard (Report, August 2024) (‘Voluntary AI Safety Standard’).

^[7] Proposed Mandatory Guardrails (n 4).

^[8] EU AI Act (n 5): see in particular arts 13, 18 and 50.

^[9] Australia’s AI Ethics Principles (n 2).

^[10] Voluntary AI Safety Standard (n 6) 32. See also iv, vi.

^[11] Proposed Mandatory Guardrails (n 4) 30.

^[12] Voluntary AI Safety Standard (n 6) 32.

^[13] See also Transparency Requirements on Content Moderation Applying to Digital Platforms in Regulation (EU) 2022/2065 of the European Parliament and of the Council of 19 October 2022 on a Single Market for Digital Services and amending Directive 2000/31/EC (Digital Services Act) [2022] OJ L 277/1 (‘Transparency Requirements on Content Moderation Applying to Digital Platforms in Regulation’).

^[14] See Cathy O'Neil, Holli Sargeant and Jacob Appel, ‘Explainable Fairness in Regulatory Algorithmic Auditing’ (2024) 127 West Virginia Law Review (forthcoming).

^[15] EU AI Act (n 5): see in particular recitals 20 and 27.

^[16] Jeannie Paterson, ‘Misleading AI: Regulatory Strategies for Algorithmic Transparency in Technologies Augmenting Consumer Decision-Making’ (2023) 34(3) Loyola Consumer Law Review 1, 13.

^[17] Competition and Consumer Act 2010 (Cth) sch 2, s 18 (‘Australian Consumer Law’).

^[18] See especially Miller & Associates Insurance Broking Pty Ltd v BMW Australia Finance Ltd (2010) 241 CLR 357, 368–9 [16]–[21] (French CJ and Kiefel J).

^[19] Jeannie Marie Paterson and Elise Bant, ‘Intuitive Synthesis and Fidelity to Purpose? Judicial Interpretation of the Discretionary Power to Award Civil Penalties under the Australian Consumer Law’ in Prue Vines and Donald M Scott (eds), Statutory Interpretation in Private Law (Federation Press, 2019) 154.

^[20] Proposed Mandatory Guardrails (n 4) 4.

^[21] See also Jeannie Paterson and Yvette Maker ‘Consumer Protection Law and AI’, in Ernest Lim and Phillip Morgan (eds), The Cambridge Handbook of Private Law and Artificial Intelligence (Cambridge University Press, 2024).

^[22] Australian Consumer Law (n 17) pt 3 ch 2.

^[23] Voluntary AI Safety Standard (n 6).

^[24] See, eg, Jeannie Marie Paterson, ‘Making Robo-advisers Careful? Duties of Care in Providing Automated Financial Advice to Consumers’ (2021) 15(3–4) Law and Financial Markets Review 278.

^[25] Frank Pasquale, The New Laws of Robotics (Belknap Press, 2020) 7.

^[26] See Judith Shulevitz, ‘Alexa, How Will You Change Us?’, The Atlantic Magazine (online, 15 November 2018) <www.theatlantic.com/magazine/archive/2018/11/alexa-how-will-you-change-us/570844/>.

^[27] Cf. ‘Principles of Robotics’, Engineering and Physical Sciences Research Council (Web Page) <https://webarchive.nationalarchives.gov.uk/ukgwa/20210701125353/https://epsrc.ukri.org/research/ourportfolio/themes/engineering/activities/principlesofrobotics/> (identifying in modern laws of robotics that the robot will not deceive humans as to its identity).

^[28] See Jeannie Marie Paterson, ‘AI Mimicking Humans’ (2024) Journal of Bioethical Inquiry (forthcoming).

^[29] Joanna J Bryson, ‘Robots Should Be Slaves’ in Yorick Wilks (ed) Close Engagements with Artificial Companions: Key Social, Psychological, Ethical and Design Issues (John Benjamins Publishing, 2010) 63, 70–2; Simon Chesterman, We the Robots? (Cambridge University Press, 2021) 114. Cf. John Danaher, ‘The Philosophical Case for Robot Friendship’ (2019) 3(1) Journal of Posthuman Studies 5.

^[30] Cal Bus & Prof Code § 17941 (2023).

^[31] EU AI Act (n 5) art 50.

^[32] Voluntary AI Safety Standard (n 6) 32. See also iv and vi.

^[33] Royal Commission into the Robodebt Scheme (Report, 7 July 2023) vol 1–3.

^[34] See generally Janina Boughey and Katie Miller (eds), The Automated State (Federation Press, 2021).

^[35] Michael Atkin, ‘Insurance Giant IAG Facing Class Action for Allegedly Inflating Premiums of Loyal Customers Across Australia’, Australian Broadcasting Corporation (online, 29 May 2024) <https://www.abc.net.au/news/2024-05-29/iag-insurance-class-action-for-inflating-loyal-customers-bills/103884968>.

^[36] On the scope of the Australian Consumer Law: see Jeannie Paterson, Corones’ Australian Consumer Law (Law Book Co, 5th ed, 2023) [2.190]–[2.200].

^[37] See Zofia Bednarz and Monika Zalnieriute (eds), Money, Power, and AI (Cambridge University Press, 2023).

^[38] Privacy and Other Legislation Amendment Bill 2024 (Cth) pt 15.

^[39] Regulation (EU) 2016/679 of the European Parliament and of the Council of 27 April 2016 on the Protection of Natural Persons with Regard to the Processing of Personal Data and on the Free Movement of such Data, and Repealing Directive 95/46/EC [2016] OJ L 119/1, art 22.

^[40] Proposed Mandatory Guardrails (n 4).

^[41] eSafety Commissioner, Deepfakes | What are Deepfakes? (Web Page, 18 March 2024) <https://www.esafety.gov.au/industry/tech-trends-and-challenges/deepfakes>.

^[42] Hannah Smith and Katherine Mansted, Australian Strategic Policy Institute, Weaponised Deep Fakes (Policy Brief No 28/2020, 29 April 2020).

^[43] Roman Kepczyk, ‘Deepfakes Emerge as Real Cybersecurity Threat’, American Institute of Certified Public Accountants & Chartered Institute of Management Accountants (online, 28 September 2022) <https://www.aicpa-cima.com/news/article/deepfakes-emerge-as-real-cybersecurity-threat>; ‘Deepfake Colleagues Trick HK Clerk into Paying HK $200m’, Radio Television Hong Kong (online, 4 February 2024) <https://news.rthk.hk/rthk/en/component/k2/1739119-20240204.htm>.

^[44] Danielle Citron and Robert Chesney, ‘Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security’ (2019) 107(6) California Law Review 1753, 1771.

^[45] Emine Saner, ‘Inside the Taylor Swift Deepfake Scandal: “It’s Men Telling a Powerful Woman to Get Back in Her Box”’, The Guardian (online, 31 January 2024) <https://www.theguardian.com/technology/2024/jan/31/inside-the-taylor-swift-deepfake-scandal-its-men-telling-a-powerful-woman-to-get-back-in-her-box>.

^[46] Australian Associated Press, ‘Australian Mining Magnate Andrew Forrest Can Sue Meta Over Facebook Scam Ads, US Court Rules’, The Guardian (online, 19 June 2024) <https://www.theguardian.com/technology/article/2024/jun/19/metas-bid-to-dismiss-case-brought-by-andrew-forrest-over-facebook-scam-ads-dismissed-by-us-court>.

^[47] Editorial, ‘The Guardian View on Political Deepfakes: Voters Can’t Believe Their Own Eyes’, The Guardian (online, 20 February 2024) <https://www.theguardian.com/commentisfree/2024/feb/19/the-guardian-view-on-political-deepfakes-voters-cant-believe-their-lying-eyes>.

^[48] Citron and Chesney (n 44) 1785.

^[49] Criminal Code Amendment (Deepfake Sexual Material) Bill 2024 (Cth). Also See generally Asher Flynn, Jonathan Clough and Talani Cooke, ‘Disrupting and Preventing Deepfake Abuse: Exploring Criminal Law Responses to AI-Facilitated Abuse’ in Anastasia Powell, Asher Flynn and Lisa Sugiura (eds), The Palgrave Handbook of Gendered Violence and Technology (Palgrave Macmillan, 2021).

^[50] eSafety Commissioner, Regulatory Schemes (Web Page, 11 March 2024) <www.esafety.gov.au/about-us/who-we-are/regulatory-schemes>.

^[51] See also Transparency Requirements on Content Moderation Applying to Digital Platforms in Regulation (n 13).

^[52] Communications Legislation Amendment (Combating Misinformation and Disinformation) Bill 2024 (Cth).

^[53] Australian Government Treasury, Scams Prevention Framework: Summary of Reforms (Summary Report, September 2024).

^[54] See generally Explanatory Materials, Treasury Laws Amendment Bill 2024: Scams Prevention Framework.

^[55] The White House, ‘Fact Sheet: Biden-⁠Harris Administration Secures Voluntary Commitments from Eight Additional Artificial Intelligence Companies to Manage the Risks Posed by AI’ (Media Release, 12 September 2023) <https://www.whitehouse.gov/briefing-room/statements-releases/2023/09/12/fact-sheet-biden-harris-administration-secures-voluntary-commitments-from-eight-additional-artificial-intelligence-companies-to-manage-the-risks-posed-by-ai/>.

^[56] ‘C2PA Technical Specification 1.3’, Coalition for Content Provenance and Authenticity (Technical Standards Framework, 29 March 2023) <https://c2pa.org/specifications/specifications/1.3/specs/_attachments/C2PA_Specification.pdf>.

^[57] See Siddarth Srinivasan, ‘Detecting AI Fingerprints: A Guide to Watermarking and Beyond’ (Research Paper, The Brookings Institution, 4 January 2024) <https://www.brookings.edu/articles/detecting-ai-fingerprints-a-guide-to-watermarking-and-beyond/>. See also Melissa Heikkilä, ‘Three Ways We Can Fight Deepfake Porn’ (29 January 2024) MIT Technology Review.

^[58]‘C2PA Technical Specification’, Coalition for Content Provenance and Authenticity (Technical Standards Framework, 26 January 2022) <https://c2pa.org/specifications/specifications/2.1/specs/C2PA_Specification.html>.

^[59] Emilia David, ‘Watermarking the Future’, The Verge (online, 14 February 2024) <https://www.theverge.com/2024/2/13/24067991/watermark-generative-ai-deepfake-copyright>.

^[60] Kathryn Harrison and Amelia Leopold, ‘How Blockchain Can Help Combat Disinformation’, Harvard Business Review (online, 19 July 2021) <https://hbr.org/2021/07/how-blockchain-can-help-combat-disinformation >.

^[61] Vittoria Elliot, ‘Big AI Won’t Stop Election Deepfakes with Watermarks’, Wired (online, 27 July 2023).

^[62] On the free speech concerns: see Lorraine Finlay, ‘Why Misinformation Bill Risks Freedoms it Aims to Protect’, Australian Human Rights Commission (online, 24 August 2023) <https://humanrights.gov.au/about/news/opinions/why-misinformation-bill-risks-freedoms-it-aims-protect>. John Storey, Institute of Public Affairs, ‘Revised Misinformation Laws Amp Up Assault on Free Speech’ (Media Release, 12 September 2024) <https://ipa.org.au/publications-ipa/media-releases/revised-misinformation-laws-amp-up-assault-on-free-speech>.

^[63] See Jeannie Marie Paterson, ‘So You Have Been Scammed by a Deepfake. What Do You Do?’, The Conversation (online, 26 February 2024) <https://theconversation.com/so-youve-been-scammed-by-a-deepfake-what-can-you-do-223299>.