In January 2024, sexually explicit deepfake images of Taylor Swift circulated on X (formerly Twitter), with one image reportedly viewed over forty-seven million times before removal (Specia, 2024). The episode crystallised a problem that English law has been slow to confront: the non-consensual fabrication of hyperrealistic sexual imagery using generative artificial intelligence. The Online Safety Act 2023 criminalised the sharing of such material, and the previous Government announced in April 2024 an intention to criminalise its creation (Ministry of Justice, 2024). The current Government has since signalled, through the Crime and Policing Bill 2025, that creation will indeed be made a criminal offence. The doctrinal and normative question, however, is not whether to criminalise, but how. Should the creation of sexually explicit deepfakes be designated a “serious sexual offence” — meaning, in the English context, an indictable sexual offence carrying significant custodial sanctions, notification requirements under Part 2 of the Sexual Offences Act 2003, and the procedural and evidential protections of the Youth Justice and Criminal Evidence Act 1999?
This essay argues that creation of sexually explicit deepfakes should be treated as a serious sexual offence, but only where the conduct displays the moral architecture of a sexual wrong: namely, where the image is of an identifiable person, is created without consent, and is created with the purpose of sexual gratification, humiliation, alarm or distress, or for distribution. Consequently, a blanket criminalisation untethered from these elements risks both over-criminalisation of trivial or artistic uses and, paradoxically, under-protection of victims because juries and courts will resist convicting under an overbroad rule. The stronger model treats the wrong as a violation of sexual autonomy and dignity analogous to voyeurism and image-based abuse, rather than as a property-style harm to reputation or a generic communications offence. Consequently, the offence should sit within the architecture of the Sexual Offences Act 2003, attract notification requirements, and benefit from the special measures available to complainants in sexual cases.
The conceptual stakes: what makes a sexual offence “serious”?
The question presupposes a workable threshold for the category “serious sexual offence”. English law does not deploy this label with doctrinal precision, but it carries identifiable institutional consequences. An offence is treated as a serious sexual offence where it is triable on indictment, carries substantial custodial maximums, triggers automatic notification requirements under Part 2 of the Sexual Offences Act 2003 (the “sex offender register”), engages the special measures regime under sections 16–33 of the Youth Justice and Criminal Evidence Act 1999, and ordinarily attracts complainant anonymity under the Sexual Offences (Amendment) Act 1992. Consequently, classifying deepfake creation as a serious sexual offence is not a rhetorical gesture; it imports a substantial procedural and regulatory apparatus.
The normative question is therefore whether the wrong of creating a sexually explicit deepfake is sufficiently analogous to other conduct attracting that apparatus. McGlynn and Rackley’s influential work on image-based sexual abuse argues that non-consensual sexual imagery — whether genuine or fabricated — constitutes a continuum of sexual violence, defined by the violation of sexual autonomy and dignity rather than by physical contact (McGlynn and Rackley, 2017). On this view, the harm of a deepfake is not derivative of defamation, harassment or copyright; it is the harm of being sexually objectified and used without consent. Consequently, the wrong belongs alongside voyeurism (section 67, Sexual Offences Act 2003) and the non-consensual sharing of intimate images (now consolidated in section 66B, Sexual Offences Act 2003, as inserted by the Online Safety Act 2023).
This characterisation is contested. Rowbottom has argued more cautiously that criminalisation of synthetic speech raises distinctive expressive concerns and that prohibitions must be carefully calibrated (Rowbottom, 2018). The point is well taken, but it cuts against overbroad drafting rather than against treating the core wrong as sexual. The position defended here is that the conceptual fit with sexual offending is strong where, and only where, the four conditions identified above are met.
The existing legal landscape and its gaps
Until very recently, English criminal law addressed sexually explicit deepfakes only obliquely. Section 33 of the Criminal Justice and Courts Act 2015 criminalised disclosure of “private sexual photographs and films” but, as Gillespie observed early in the debate, the requirement that the image be a “photograph or film” of a real event excluded fabricated imagery (Gillespie, 2019). The Law Commission’s 2022 report on intimate image abuse confirmed the gap, recommending an expanded regime covering altered and wholly fabricated images (Law Commission, 2022).
Parliament responded through section 188 of the Online Safety Act 2023, which inserted a new section 66B into the Sexual Offences Act 2003. The new section criminalises sharing, and threatening to share, intimate photographs or films, with the definition in section 66D expressly including images that “appear to show” the person and images that have been “altered”. Consequently, distribution of a sexually explicit deepfake of an identifiable adult is now an offence, with the base offence triable either way and aggravated variants where the purpose is to cause alarm, distress or humiliation, or for sexual gratification.
Creation, however, was not covered. The Ministry of Justice announced in April 2024 an intention to criminalise creation, and the Crime and Policing Bill introduced in early 2025 contains provisions to that effect. The drafting matters intensely. The 2024 announcement initially indicated a non-imprisonable offence carrying an unlimited fine, with a community order available; this is markedly less serious than the corresponding distribution offences. Consequently, the question posed by this essay is live: even assuming creation is criminalised, should it be treated as a serious sexual offence, with the full apparatus that designation imports?
The voyeurism analogy and its limits
Section 67 of the Sexual Offences Act 2003 criminalises observing or recording another doing a “private act” for sexual gratification without consent. Section 67A, inserted by the Voyeurism (Offences) Act 2019 following the campaign led by Gina Martin, extended the offence to “upskirting”. Both offences are triable either way, carry a maximum of two years’ imprisonment, and — crucially — trigger notification requirements where committed for sexual gratification (Schedule 3, Sexual Offences Act 2003). The analogy with deepfake creation is direct: in each case, the offender appropriates a sexualised representation of another without consent, and the wrong is constituted by that appropriation independently of any subsequent dissemination.
The analogy is not perfect. Voyeurism requires a “private act” actually taking place; deepfake creation fabricates the appearance of such an act. One might therefore argue that the deepfake offender invades no actual privacy. The response, drawing on McGlynn, Rackley and Houghton (2017), is that the protected interest is sexual autonomy and dignitary control over one’s sexualised image, not a narrow informational privacy. The deepfake victim experiences essentially the same harm as the voyeurism victim — a non-consensual sexual representation of herself exists, can be viewed, can be circulated — with the aggravating feature that the representation purports to depict acts she never performed. Citron’s work on intimate privacy and Franks’s analysis of “fake porn” both emphasise that the harm of synthetic imagery is, if anything, more acute because the victim cannot disprove the image to every viewer (Citron, 2019; Franks and Waldman, 2019). Consequently, treating creation less seriously than voyeurism would be doctrinally inconsistent.
The case for treating creation as a serious sexual offence
The wrong is the violation of sexual autonomy, not the dissemination
The strongest argument for serious classification is that the relevant wrong crystallises at creation, not at distribution. Once a sexually explicit deepfake exists, the victim’s sexual autonomy has been violated in a manner that is not contingent on any third party seeing the image. The offender has appropriated her likeness to fabricate a sexual representation he could not lawfully have obtained from her. That is the same structural wrong as recording a person without consent: the harm is the existence of the sexualised representation, the possibility of its circulation, and the psychological reality that the victim cannot recall it.
This characterisation explains why purely private creation, where no further conduct is intended, remains seriously wrongful. The point is reinforced by empirical evidence on victim experience. McGlynn et al.’s qualitative research with victims of image-based abuse documents profound and lasting harm — including suicidal ideation, withdrawal from employment, and chronic anxiety — that is not dependent on the scale of dissemination but on the fact of non-consensual sexualisation (McGlynn et al., 2021). Consequently, an offence that captures only distribution treats the wrong as derivative; an offence that captures creation recognises it as primary.
The Taylor Swift episode demonstrates the scale and the speed
The January 2024 incident is instructive not because high-profile victims deserve greater protection but because it illustrates the structural features of the harm. The images were generated using widely available generative tools, reportedly via the “Designer” image generator after users discovered prompt circumventions (Belanger, 2024). They were then shared on a major platform and viewed at scale before moderation responded. The case demonstrates three points relevant to legal design.
First, the harm propagates faster than enforcement. By the time distribution offences engage, the image has already been viewed millions of times. Consequently, a criminal regime targeting only distribution accepts that the primary harm will routinely have occurred before the law bites. Second, the technical barrier is now negligible: any user with consumer hardware can produce a convincing deepfake. Consequently, deterrence must operate at the point of creation. Third, the victim class is not confined to celebrities. Internet Matters and other survey work indicates that schoolchildren are increasingly the victims of peer-generated deepfakes (Internet Matters, 2025), and reporting by the Internet Watch Foundation has documented an explosion in AI-generated child sexual abuse material (Internet Watch Foundation, 2023). The adult-image offence and the existing child sexual abuse image offences under the Protection of Children Act 1978 form, on the better view, parts of a single regulatory architecture, and the adult limb must be calibrated to similar seriousness.
Doctrinal coherence with the Sexual Offences Act 2003
If creation is criminalised but not designated as a sexual offence, the result is doctrinally incoherent. Distribution of the same image under section 66B is a sexual offence, attracts notification where committed for sexual gratification, and engages complainant anonymity. The creator who fabricates an image but does not distribute it would, on the original 2024 proposal, commit a non-imprisonable summary offence. Consequently, the more culpable conduct — the manufacture of the harmful artefact — would attract less serious classification than the downstream act of sharing. That inversion is difficult to defend on any plausible theory of culpability. Hörnle’s comparative work on sexual offence categorisation argues that the gravamen of sexual offending lies in the violation of sexual self-determination, and that legal systems achieve coherence by tracking that violation through the chain of conduct (Hörnle, 2014). Consequently, creation, which initiates the violation, should not be graded below distribution, which extends it.
Notification requirements and risk management
Notification under Part 2 of the Sexual Offences Act 2003 is not a punitive sanction but a risk-management measure. Its application to deepfake creators is justifiable where the offence is committed for sexual gratification, because the conduct evidences a sexual interest in non-consenting persons that is relevant to risk assessment. The same logic underpins notification for section 67 voyeurism committed for sexual gratification. Thomas’s work on the sex offender register notes that its legitimacy depends on a defensible nexus between the index offence and ongoing risk (Thomas, 2011). Where the deepfake offence is created with sexual purpose and targets identifiable individuals, that nexus is established. Consequently, notification is appropriate, though it should be calibrated by the purpose element rather than applied automatically.
The case against serious classification — and why it does not prevail
The expressive and over-criminalisation objection
The most serious objection is that creation is closer to private thought or fantasy than to publication, and that criminalising creation invades a domain ordinarily reserved to autonomous expression. Rowbottom’s analysis of synthetic speech is relevant: he warns that the conceptual elasticity of “harm” in the online sphere may permit criminal law to encroach on imagination and private creative activity (Rowbottom, 2018). The objection has weight where the conduct is genuinely private, non-targeted, or expressive (for example, satire, art or pornography that uses synthetic faces not corresponding to identifiable individuals).
The response is that the proposed offence does not criminalise imagination; it criminalises the manufacture of a sexual representation of an identifiable real person without consent. That is conduct, not thought. The expressive concern is real but is addressed by carefully drawn elements: the image must be of an identifiable person, the person must not have consented, and the offender must have acted with a defined culpable purpose. Consequently, the offence should not extend to fully synthetic imagery not depicting an identifiable individual, nor to private artistic or political expression that does not satisfy the purpose element. Properly drafted, the offence is narrower than section 127 of the Communications Act 2003 or section 1 of the Malicious Communications Act 1988, both of which have attracted longstanding criticism for vagueness.
The mens rea problem
A related objection is evidential: how can prosecutors prove that the offender knew the image was non-consensual, or acted with sexual purpose? The Crown Prosecution Service’s experience with section 66B and its predecessor in section 33 of the 2015 Act demonstrates the difficulties. McGlynn and Rackley have criticised the requirement of intent to cause distress in section 33 as setting too high a bar, with prosecutors unable to charge cases where the offender acted “for a laugh” or for sexual gratification (McGlynn and Rackley, 2017). The 2023 reform partially addressed this by creating a base offence in section 66B(1) that does not require proof of motive, with aggravated variants where particular purposes are proved.
Consequently, the deepfake creation offence should adopt a similar tiered structure: a base offence of intentional creation of a sexually explicit image of an identifiable person without consent, and aggravated variants where the creation is for sexual gratification or for the purpose of distribution. Only the aggravated variants need attract notification requirements. This design preserves the seriousness of the offence while addressing the prosecutorial bottleneck identified in the post-2015 experience.
The platform displacement objection
A pragmatic objection is that criminal law cannot effectively address a problem driven by platform design and the architecture of generative AI services. Suzor’s work on platform governance suggests that the most effective interventions are upstream, at the level of model training, content filtering and platform liability (Suzor, 2019). The point is correct but not a counter-argument. The Online Safety Act 2023 imposes duties on regulated services in respect of illegal content, and the designation of deepfake creation as a serious sexual offence has direct downstream consequences for platform duties under section 59 of that Act, because “priority illegal content” includes sexual offences. Consequently, treating creation as a serious sexual offence is not an alternative to platform regulation; it is the trigger for it.
The disproportionality concern: schoolchildren and young offenders
A genuinely difficult objection is that designating creation as a serious sexual offence will catch young offenders, including children at school producing deepfakes of classmates. The data suggest this is not a hypothetical concern: Internet Matters’ 2025 survey reports that significant numbers of children encounter or produce deepfake nudes in peer contexts (Internet Matters, 2025). Notification requirements applied to a fifteen-year-old offender raise serious proportionality concerns under Article 8 of the European Convention on Human Rights, as the Supreme Court recognised in R (F) v Secretary of State for the Justice [2010] UKSC 17 in respect of indefinite notification.
The objection does not, however, defeat serious classification. It defeats automatic application of all consequences of that classification to all offenders. The appropriate response is a calibrated regime in which the substantive offence and its maximum sentence reflect the seriousness of the wrong, while the application of notification requirements and other ancillary measures is governed by judicial discretion, particularly in youth cases. The youth justice system already mediates the application of sexual offence consequences to young offenders through the Youth Court and through specialist disposals. Consequently, the proportionality concern is properly addressed at the level of sentencing and ancillary order, not at the level of substantive classification.
The shape of a defensible offence
The argument so far supports treating creation as a serious sexual offence. The remaining question is how the offence should be drafted to achieve that without the pathologies of overbreadth.
Element 1: identifiable person
The image must depict an identifiable real person. This excludes wholly synthetic imagery, animated content not corresponding to a real individual, and content where the depicted person is not reasonably identifiable. The Law Commission’s 2022 report endorsed this requirement and it tracks the structure of section 66B (Law Commission, 2022). Consequently, the offence does not extend to imagination or to expression that uses no real person’s likeness.
Element 2: sexual content
The image must be sexually explicit, defined consistently with section 66D of the Sexual Offences Act 2003 as inserted by the Online Safety Act 2023. Consistency across the creation and distribution offences is important to avoid the kind of definitional gap that bedevilled the pre-2023 regime.
Element 3: absence of consent
The image must be created without the depicted person’s consent. Consent must be defined consistently with section 74 of the Sexual Offences Act 2003 — agreement by choice with the freedom and capacity to make that choice. Consent to the underlying photograph (for example, a clothed image used as the source) does not constitute consent to its sexualisation. This is not a controversial extension; it tracks the existing approach in section 66B and reflects the Law Commission’s recommended framework.
Element 4: culpable purpose, in a tiered structure
The base offence should require intentional creation knowing of the absence of consent, with no further purpose required. The aggravated offence should require an additional purpose: sexual gratification, causing alarm, distress or humiliation, or onward distribution. Only the aggravated offence should attract notification requirements. This tiered model resolves the central tension between under-enforcement (where prosecutors must prove motive in a private act) and over-punishment (where notification attaches to relatively minor offending). It also mirrors the structure adopted by Parliament for the distribution offence in section 66B, achieving doctrinal coherence.
Defences
Limited defences should be available: reasonable belief in consent (consistent with the approach taken to consent across the Sexual Offences Act 2003); legitimate purposes such as law enforcement, evidence preservation, or bona fide journalistic investigation of deepfake abuse; and use in legal proceedings. The case for an artistic or satirical defence is weaker. Where art or satire involves the sexual depiction of an identifiable real person without consent, the expressive interest does not outweigh the dignitary harm; Article 10 ECHR jurisprudence consistently affords lower protection to such expression where it engages others’ Article 8 rights, as in Von Hannover v Germany (No 2) (2012) 55 EHRR 15.
Comparative perspective: a brief and selective view
Several jurisdictions have moved faster than England and Wales. South Korea’s revised Sexual Violence Punishment Act criminalises the production of “false sexual images” with custodial penalties, and prosecutions have followed (though the legislation has also attracted criticism for definitional vagueness). In the United States, federal legislation has been proposed but, at the time of writing, the regulatory response has been primarily at state level, with variable definitions. The European Union’s AI Act imposes transparency obligations on deepfake generators (Article 50) but does not itself create criminal offences for non-consensual sexual deepfakes; criminal law remains a Member State competence. Consequently, comparative work supports the view that criminalisation is converging across jurisdictions, but the design choices vary substantially. The English approach should not be designed by transplant; it should reflect the architecture of the Sexual Offences Act 2003 and the Online Safety Act 2023, into which the creation offence must integrate.
The procedural consequences that designation imports
Treating creation as a serious sexual offence has consequences beyond sentencing maximums. Three are particularly important.
First, complainant anonymity under the Sexual Offences (Amendment) Act 1992 would attach automatically. This matters because complainants of deepfake abuse have repeatedly described the prospect of being publicly identified as a primary deterrent to reporting (McGlynn et al., 2021). Consequently, anonymity is not a procedural luxury; it is a precondition for the offence functioning at all.
Second, special measures under sections 16–33 of the Youth Justice and Criminal Evidence Act 1999 would be available, including screens, live link evidence, and video-recorded evidence-in-chief. The need is acute: viewing the image during testimony may itself constitute a substantial harm.
Third, the offence would be designated as priority illegal content under the Online Safety Act 2023, engaging the safety duties on regulated services. This is the principal mechanism by which the criminal classification produces upstream prevention. Consequently, the choice to designate creation as a serious sexual offence is, in practical terms, also a choice about the scope of platform regulation.
Sentencing, deterrence and the limits of criminal law
A frequent objection to robust criminalisation is that deterrence is limited where offenders are anonymous, jurisdictionally distant, or technically sophisticated. The point is partly correct: criminal law alone will not solve the deepfake problem. However, deterrence is not the only function of criminal classification. Expressive and communicative functions also matter. Duff’s communicative theory of punishment holds that the criminal law’s role is partly to communicate the seriousness with which the polity regards a wrong (Duff, 2007). Consequently, classifying deepfake creation as a serious sexual offence performs essential expressive work even if marginal deterrence is modest. It communicates to victims that the harm is recognised, to potential offenders that the conduct is taken seriously, and to platforms that their content moderation must reflect that gravity.
The sentencing maximums should reflect the structure of section 66B: two years’ imprisonment for the base offence on indictment, with potentially higher maximums for the aggravated variants. The original 2024 proposal of an unlimited fine and community order was insufficient and would have created the inversion noted above. The Crime and Policing Bill 2025 reportedly contemplates more serious penalties; the precise drafting will determine whether the resulting regime is coherent.
Counterargument revisited: the chilling effect on legitimate AI development
One final objection deserves direct address. Industry commentators have argued that criminalising creation will chill legitimate AI research and development, because researchers may produce sexually explicit synthetic outputs as artefacts of model development or red-teaming. The objection misunderstands the offence. As proposed, the offence requires that the image depict an identifiable real person without consent. Legitimate red-teaming and safety research does not require depicting identifiable individuals. Consequently, the offence does not bite on bona fide research, and a researcher engaged in such work can rely on the absence of any element being satisfied. The chilling-effect argument therefore fails on its own terms.
Conclusion
The creation of sexually explicit deepfakes should be treated as a serious sexual offence. The wrong is a violation of sexual autonomy and dignity that crystallises at the moment of creation, before any distribution. It is structurally analogous to voyeurism and to the non-consensual sharing of intimate images, both of which are properly treated as sexual offences attracting notification requirements where committed for sexual gratification. Consequently, doctrinal coherence with the Sexual Offences Act 2003 — and particularly with section 66B as inserted by the Online Safety Act 2023 — requires that creation not be graded below distribution.
The objection that serious classification will over-criminalise legitimate expression has force only against overbroad drafting. A properly drafted offence requiring an identifiable victim, absence of consent and (for the aggravated variant) a culpable purpose excludes imagination, satire of public events, fully synthetic imagery and bona fide research. The objection that the regime will catch young offenders disproportionately is genuine but is properly addressed through sentencing discretion and youth-justice mediation of ancillary consequences, not through dilution of the substantive offence.
Consequently, the recommended position is a tiered offence sitting within the Sexual Offences Act 2003: a base offence of intentional creation of a sexually explicit image of an identifiable adult without consent, and an aggravated offence where the creation is for sexual gratification, for the purpose of causing alarm, distress or humiliation, or for distribution, with notification attaching only to the aggravated variant. That design reflects the gravity of the wrong, integrates with the existing legislative architecture, addresses the institutional consequences identified, and limits the over-criminalisation concerns that have rightly attracted scholarly attention. The Taylor Swift episode is unlikely to be the last of its kind, and the schoolchildren whose images are circulated in peer contexts deserve the same recognition. The question is not whether the criminal law should respond, but whether the response will be doctrinally serious. On the better view, it should be.
References
- Belanger, A. (2024) ‘Taylor Swift deepfakes on X linked to Microsoft Designer’s “loophole”‘. Ars Technica, 30 January.
- Citron, D.K. (2019) ‘Sexual privacy’. Yale Law Journal, 128(7), pp. 1870–1960.
- Criminal Justice and Courts Act 2015, s 33.
- Duff, R.A. (2007) Answering for Crime: Responsibility and Liability in the Criminal Law. Oxford: Hart Publishing.
- Franks, M.A. and Waldman, A.E. (2019) ‘Sex, lies, and videotape: deep fakes and free speech delusions’. Maryland Law Review, 78(4), pp. 892–898.
- Gillespie, A.A. (2019) ‘”Trust me, it’s only for me”: “revenge porn” and the criminal law’. Criminal Law Review, 2019(10), pp. 866–880.
- Hörnle, T. (2014) ‘Theories of criminalization’, in Dubber, M.D. and Hörnle, T. (eds) The Oxford Handbook of Criminal Law. Oxford: Oxford University Press.
- Internet Matters (2025) The new face of digital abuse: Children’s experiences of nude deepfakes. London: Internet Matters.
- Internet Watch Foundation (2023) How AI is being abused to create child sexual abuse imagery. Cambridge: IWF.
- Law Commission (2022) Intimate Image Abuse: A Final Report. Law Com No 407. London: HMSO.
- McGlynn, C. and Rackley, E. (2017) ‘Image-based sexual abuse’. Oxford Journal of Legal Studies, 37(3), pp. 534–561.
- McGlynn, C., Rackley, E. and Houghton, R. (2017) ‘Beyond “revenge porn”: the continuum of image-based sexual abuse’. Feminist Legal Studies, 25(1), pp. 25–46.
- McGlynn, C., Johnson, K., Rackley, E., Henry, N., Gavey, N., Flynn, A. and Powell, A. (2021) ‘”It’s torture for the soul”: the harms of image-based sexual abuse’. Social and Legal Studies, 30(4), pp. 541–562.
- Ministry of Justice (2024) Government cracks down on “deepfakes” creation. Press release, 16 April. London: Ministry of Justice.
- Online Safety Act 2023.
- Protection of Children Act 1978.
- R (F) v Secretary of State for the Home Department [2010] UKSC 17.
- Rowbottom, J. (2018) Media Law. Oxford: Hart Publishing.
- Sexual Offences Act 2003.
- Sexual Offences (Amendment) Act 1992.
- Specia, M. (2024) ‘Explicit deepfake images of Taylor Swift elude safeguards and swamp social media’. The New York Times, 26 January.
- Suzor, N. (2019) Lawless: The Secret Rules That Govern Our Digital Lives. Cambridge: Cambridge University Press.
- Thomas, T. (2011) The Registration and Monitoring of Sex Offenders: A Comparative Study. Abingdon: Routledge.
- Von Hannover v Germany (No 2) (2012) 55 EHRR 15.
- Voyeurism (Offences) Act 2019.
- Youth Justice and Criminal Evidence Act 1999.
