The Fiction of Consent: AI-Generated Sexual Deepfakes and the Limits of Australian Law

Each article has been preserved in its original submitted form. No edits have been made to grammar, spelling or AGLC4 referencing.

By Nithya Appanah

I INTRODUCTION

At eighteen, an Australian law student ran a reverse-image search on a photograph of herself. Noelle Martin found her face grafted onto pornographic images – and later, into deepfake videos – depicting acts that never occurred.[1] No offence fitted what had been done to her: the images were not ‘real’, so the law treated the harm as unreal too.

Generative AI has since made such abuse effortless, and Australia’s response has been a patchwork. Victoria criminalised digitally creating intimate images from July 2023;[2] the Commonwealth prohibited transmitting them in 2024;[3] New South Wales left creating an adult sexual deepfake outside its criminal law until 16 February 2026.[4] This essay argues that the patchwork is indefensible – for years, a victim’s protection turned on her postcode, for abuse overwhelmingly directed at women – and that the deeper defect cannot be patched at all. Statutes built around camera-made recordings must be rebuilt around the harm: non-consensual sexualisation, not the provenance of pixels. A deepfake trades on a fiction of consent – a willing participant who never consented. The law indulged a second fiction: that a fabricated image inflicts only a fabricated harm.

II REAL HARM FROM FAKE IMAGES

Scholars conceptualise the non-consensual creation and distribution of sexual imagery as ‘image-based sexual abuse’: a form of sexual violence on a continuum with other sexual offences, violating sexual autonomy, dignity and privacy.[5] Deepfakes are its fastest-growing and most gendered form. The first systematic audit of online deepfakes found 96 per cent were non-consensual sexual videos, and 99 per cent of those depicted women;[6] by 2023, a further audit counted 95,820 such videos – a 550 per cent increase in four years – of which 98 per cent were pornographic and 99 per cent targeted women.[7] Nor is the abuse confined to celebrities: in a survey across Australia, the United Kingdom and New Zealand, 14.1 per cent of respondents reported victimisation involving deepfake or digitally altered sexual imagery.[8] Victims of fabricated imagery describe harms indistinguishable from those of ‘real’ recordings: shame, hypervigilance, withdrawal from public and online life, damage to employment and relationships.[9] That should be unsurprising. What injures a victim is not the mechanical history of the file but her non-consensual sexualisation before others and the appropriation of her sexual identity – what Citron calls the invasion of ‘sexual privacy’.[10] Once the harm is understood this way, the image’s authenticity is beside the point. A law that protects only against ‘real’ recordings protects the evidence, not the person.

III A PATCHWORK OF PROTECTION

Victoria moved first. Its 2022 reforms, in force from 30 July 2023, made it an indictable offence, punishable by three years’ imprisonment, to produce an intimate image of another person without consent – with ‘produce’ extending to the digital creation of the image.[11] The statutory example is a deepfake in all but name: a person who ‘digitally superimposes another person’s face … onto a photograph of a naked person’.[12]

The Commonwealth followed in 2024, but along a different axis. Section 474.17A of the Criminal Code makes it an offence, punishable by six years’ imprisonment, to use a carriage service to transmit sexual material depicting an adult without consent, expressly including material created or altered using technology.[13] Where the transmitter also created the material, the offence is aggravated and carries a sentence of seven years.[14] But creation itself remains beyond the federal offence: it operates only upon transmission, a limit shaped by the constitutional reach of the carriage-service power.[15] Federally, a person may generate a photorealistic sexual image of a non-consenting woman with impunity, so long as it stays on their hard drive.

New South Wales illustrates the cost of patching. Its intimate-image offences of 2017 criminalised recording and distributing, and its definition of ‘intimate image’ included altered images – so distributing a deepfake was arguably caught.[16] However, the recording offence required that the accused ‘intentionally records’ an intimate image;[17] generating an image from scratch records nothing. Creation therefore fell into a legal grey area until the Crimes Amendment (Intimate Image and Audio Material) Act 2025 (NSW) inserted a dedicated three-year offence of producing digitally generated intimate material, commencing 16 February 2026.[18]

The result should trouble anyone committed to equality before the law. Between July 2023 and February 2026, identical conduct – generating, without sharing, a photorealistic sexual image of a non-consenting adult – was a serious indictable offence in Melbourne and no offence at all in Sydney. Victims do not choose the state in which their abuser sits at a keyboard. For a category of abuse suffered almost entirely by women,[19] Australia distributed criminal-law protection by geography: a second inequity layered upon the first.

IV ‘THE CIVIL SCHEME ALREADY PROTECTS EVERYONE’

The strongest defence of the status quo is that the criminal patchwork never mattered, because the Online Safety Act 2021 (Cth) already provides a national, harm-based scheme. Its definition of ‘intimate image’ asks only whether material ‘depicts, or appears to depict’ a person, and it is immaterial whether the material has been altered;[20] posting such an image attracts a civil penalty of up to $182,000,[21] and the eSafety Commissioner can compel removal within 24 hours.[22]

Three answers. First, the scheme is remedial, not protective: it operates only after the material exists and circulates, and only if the victim discovers it and complains. Removal cannot uncreate an image of which endless copies may persist. Second, a civil penalty mislabels the wrong. Image-based sexual abuse is a sexual violation, not a content-moderation problem; the expressive, denunciatory function of the criminal law cannot be delegated to a takedown scheme.[23] Third, and decisively, the objection concedes the thesis. The Commonwealth’s civil law has defined the wrong in terms of harm rather than authenticity since 2021. If ‘appears to depict’ suffices to make an image worth removing, the criminal law’s continued insistence on a real recording was never principled. It was inertia.

V ‘CRIMINALISING CREATION IS A THOUGHT CRIME’

Another objection to the rebuilt model runs the other way: an image generated and never shared has no audience, so criminalising creation punishes something uncomfortably close to private fantasy.

However, a deepfake is an artefact, not a thought. It persists, is infinitely copyable, and sits one upload away from mass distribution; its existence serves as a standing threat and is routinely deployed as one.[24] Fantasies cannot leak. More fundamentally, the wrong is complete at creation. The victim’s likeness – her sexual identity – has been appropriated and falsified without her consent. Sexual autonomy is not violated only when third parties watch; a woman who learns such an image exists is not comforted by its current privacy. Further, the argument that only ‘real’ depictions can ground liability is one Australian law has already abandoned where the protected interest demands it: in McEwen v Simmons, wholly fictional cartoon characters were capable of constituting child pornography.[25] The point is not to equate adult victims with children; it is doctrinal. The law already accepts that representation alone can wrong. Refusing that logic for adult women is not restraint – it is an unfinished thought. Victoria, New South Wales, and now England and Wales – whose offence of creating a purported intimate image, with no distribution requirement, commenced ten days before NSW’s – have all now legislated on this basis.[26] Residual overreach concerns are answered by drafting, not abstention: an identifiable real person, absence of consent as the gravamen, and tailored exceptions.

VI REBUILDING AROUND HARM

What would harm-based law look like? The starting point is to name the protected interest honestly: sexual autonomy and dignity, not the integrity of a recording. From there, ‘intimate image’ should be defined by what material depicts, or purports to depict, of an identifiable person, regardless of how it was made – as the Online Safety Act and NSW’s new definition now do.[27] The final piece is uniformity: creation, distribution and threats criminalised in every Australian jurisdiction on the same terms, with consent defined so that consent to one image is never consent to another.[28] As the Commonwealth’s offence hangs on the carriage-service power, it cannot reach purely local creation; genuine harmonisation therefore requires model provisions agreed through the Standing Council of Attorneys-General, as was done for defamation. The patchwork was not inevitable. It was a choice.

VII CONCLUSION

Australia reformed this field backwards: takedown first, transmission second, creation last. That ordering reveals the assumption this essay has contested – that the harm happens at publication. For the victim, the violation begins the moment the image comes into existence. For two and a half years, whether that violation was a crime depended on where the perpetrator lived, in a legal system that promises women equal protection. The lesson of the deepfake era is not that the law needs another amendment; it is that ‘authenticity’ was never the right foundation. The image was always a fiction. The harm never was.


[1] Noelle Martin, ‘Image-Based Sexual Abuse’, Noelle Martin (Web Page) <https://www.noellemartin.org/image-based-sexual-abuse>; Justin Sherman, ‘“Completely Horrifying, Dehumanizing, Degrading”: One Woman’s Fight against Deepfake Porn’, CBS News (online, 14 October 2021) <https://www.cbsnews.com/news/deepfake-porn-woman-fights-online-abuse-cbsn-originals/>.

[2] Justice Legislation Amendment (Sexual Offences and Other Matters) Act 2022 (Vic), inserting Crimes Act 1958 (Vic) ss 53O–53T.

[3] Criminal Code Act 1995 (Cth) sch 1 (‘Criminal Code’) s 474.17A, as inserted by Criminal Code Amendment (Deepfake Sexual Material) Act 2024 (Cth).

[4] Crimes Amendment (Intimate Image and Audio Material) Act 2025 (NSW); Department of Communities and Justice (NSW), ‘Law against Image-Based Abuse’, Communities and Justice (Web Page) <https://dcj.nsw.gov.au/legal-and-justice/laws-and-legislation/image-based-abuse.html>.

[5] Clare McGlynn and Erika Rackley, ‘Image-Based Sexual Abuse’ (2017) 37(3) Oxford Journal of Legal Studies 534.

[6] Henry Ajder et al, The State of Deepfakes: Landscape, Threats, and Impact (Report, Deeptrace, September 2019) 1–2.

[7] Security Hero, 2023 State of Deepfakes: Realities, Threats, and Impact (Report, 2023) <https://www.securityhero.io/state-of-deepfakes/>.

[8] Asher Flynn et al, ‘Deepfakes and Digitally Altered Imagery Abuse: A Cross-Country Exploration of an Emerging Form of Image-Based Sexual Abuse’ (2022) 62(6) British Journal of Criminology 1341.

[9] Flynn et al (n 8); McGlynn and Rackley (n 5).

[10] Danielle Keats Citron, ‘Sexual Privacy’ (2019) 128(7) Yale Law Journal 1870.

[11] Crimes Act 1958 (Vic) s 53R; Judicial College of Victoria, ‘Sexual Offence Reforms 2023’ (Web Page) <https://judicialcollege.vic.edu.au/resources/sexual-offence-reforms-2023>.

[12] Crimes Act 1958 (Vic) s 53R (example 2).

[13] Criminal Code s 474.17A.

[14] Ibid s 474.17AA(5).

[15] Explanatory Memorandum, Criminal Code Amendment (Deepfake Sexual Material) Bill 2024 (Cth).

[16] Crimes Act 1900 (NSW) ss 91N–91R (as at 1 January 2026).

[17] Crimes Act 1900 (NSW) s 91P (as at 1 January 2026).

[18] Crimes Amendment (Intimate Image and Audio Material) Act 2025 (NSW) sch 1, inserting Crimes Act 1900 (NSW) s 91PA, commencing 16 February 2026.

[19] Ajder et al (n 6); Security Hero (n 7).

[20] Online Safety Act 2021 (Cth) s 15.

[21] Ibid s 75 (500 penalty units); Crimes Act 1914 (Cth) s 4AA; Crimes (Amount of a Penalty Unit) Instrument 2026 (Cth).

[22] Online Safety Act 2021 (Cth) ss 77–79.

[23] See generally McGlynn and Rackley (n 5).

[24] Flynn et al (n 8).

[25] McEwen v Simmons (2008) 73 NSWLR 10.

[26] Crimes Act 1958 (Vic) s 53R; Crimes Act 1900 (NSW) s 91PA; Sexual Offences Act 2003 (UK) s 66E, as inserted by Data (Use and Access) Act 2025 (UK) s 138(2); Data (Use and Access) Act 2025 (Commencement No 5) Regulations 2026 (UK) SI 2026/31, reg 2.

[27] Online Safety Act 2021 (Cth) s 15; Crimes Act 1900 (NSW) s 91N, as amended by Crimes Amendment (Intimate Image and Audio Material) Act 2025 (NSW).

[28] Crimes Act 1958 (Vic) s 53P.

Share your thoughts