Home » Blog » Deepfakes and Criminal Evidence in South Africa: Challenges for the Criminal Justice System.

Deepfakes and Criminal Evidence in South Africa: Challenges for the Criminal Justice System.

Authored By: Els Dikeledi

North West University

Imagine a criminal trial where the key evidence is a video showing the accused holding a gun and threatening a witness. To the court, it looks real. To expert, it is a deepfake made in 10 minutes with free AL software.[1] In legal system build on seeing to believe, South Africa now faces a moment where seeing is no longer enough.

Deepfakes use artificial intelligence to create audio, images, or videos that are extremely realistic but completely fabricated.[2] As these tools become widely available, digital exhibits in criminal cases, confessions, intimate recordings, and threats are easier to fake and harder detect.[3] South African courts currently admit electronic evidence under section 15 of the Electronic Communication and Transactional Act and rely on expert evidence under the Criminal Procedure Act,[4] but neither framework was designed for intentional AI deception. This results in a gap between the technology in the courtroom and the law meant to regulate it.

To what extent does South African law adequately regulate the use and admissibility of digital evidence in criminal proceedings considering the challenges posed by deepfake technology?

This article argues that although South African law recognizes and regulates electronic evidence through existing statutory and common-law principles, the current legal framework is inadequate to address the evidentiary risks created by deepfake technology. The article contends that legislative reform, specialized forensic authentication techniques, and judicial guidelines are necessary to preserve the reliability of digital evidence and maintain public confidence in the criminal justice system.

Deep fake Technology

The term “deepfake” comes from “deep learning” and “fake”[5]. It refers to media created or altered with artificial intelligence, specifically generative models that can map one person’s face, voice, or mannerism onto another person’s body or speech.[6] The most common form is a face swap video, but the same tools can now clone a voice from few seconds of audio or generate an entirely synthetic person who never existed.[7] Traditional photoshop or audio splicing often left visible signs of tempering, but deep learning models are trained on thousands of images and recordings to learn how a person blinks, moves their mouths, or modulates tone.[8]  In criminal justice context, these matters because the very exhibits courts rely on most are exactly what deepfake tools target.[9]

Digital Evidence in South African Criminal Proceedings.

Digital evidence in South Africa is any evidence that is produced before the court. Digital evidence refers to any information created, stored, transmitted or retrieved in digital form that may be used to establish facts in legal proceedings.[10] This includes emails, text messages, social media communications, computer files, CCTV recordings, GPS data, and metadata generated by electronic devices.[11] Although the terms “digital evidence” and “electronic evidence” are often used interchangeably, digital evidence generally refers to information stored or processed in digital form, whereas electronic evidence encompasses all electronically generated or stored information capable of being admitted in court.[12] In South Africa, the growing reliance on digital technologies has significantly increased the use of such evidence in criminal investigations, particularly in cases involving cybercrime.[13] Consequently, courts must ensure that electronic information is collected, preserved, and presented in a manner that safeguards its integrity and reliability before it can be relied upon during criminal proceedings.[14] The legal recognition of electronic evidence is primarily governed by the Constitution of the Republic of South Africa, the Criminal Procedure Act 51 of 1977, the Electronic Communications and Transactions Act 25 of 2002, and the Cybercrimes Act 19 of 2020, which collectively provide the legal framework for the investigation, admission and evaluation of digital evidence in criminal trials.

Challenges Deepfakes presents Criminal Evidence.

Deepfakes technology does not create a new category of evidence, but it makes existing rules proving electronic evidence harder to apply in practice. In South African criminal context, four interlinked challenges stand out.

4.1 The Authentication problem.

The first difficulty is producing that a video or audio recording is what it claims to be, section 15 of the Electronic Communications and Transactions Act makes a data message admissible if the court is satisfied about its integrity and the way it was made.[15] That test assumes accidental loss of reliability, not deliberate, AI- driven fabrication. A Deepfake can be made to show a person speaking words they never said or committing an act they never did, and to the naked eye it will look authentic.[16] This shifts the burden from” is the file corrupted” to” was the file made by AI at all”, a question most magistrates and prosecutors are not equipped to answer without help.[17]

4.2 Pressure on Forensic and Expert Evidence

Because lay observation is no longer enough, courts must turn to expert witnesses.[18] South African criminal procedure already relies expert evidence to interpret electronic records under the Criminal Procedure Act and the common law.[19] The problem with Deepfakes is scale and access: detection requires specialized AI-based forensic tools and analysts trained to use them.[20] The SAPS Forensic Science Laboratory does not yet have a widespread capacity for this, which means parties with more resources will be better able to challenge or defend digital exhibits than those without.[21] That creates a real risk of inequality before the law.

4.3 Threats of the Right to Fair Trial

A convincing Deepfake can prejudice against a court even before its authenticity is tested. Section 35 (3) of the Constitution guarantees an accused the right to challenge evidence and to a fair trial.[22] When a video appears to show the accused committing the offence, it carries strong emotional and visual weight that is hard to undo, even if an expert later shows it was fabricated.[23] The opposite risk exists too: genuine recordings may be rejected simply because a party argues” it could be a Deepfake” allowing the guilty person to escape liability.[24]

4.4 Misuse against Victims and Witnesses

Beyond the courtroom, Deepfakes are already being used to intimidate or discredit people involved in criminal cases. In gender-based violence and sexual offence matters, fabricated intimate images or fake “confessions” can be circulated to silence a complainant or attract their credibility.[25] Our evidence rules do not yet account for how such material effects of witness willingness to testify or the fairness of the proceedings.[26]

Critical Evaluation of South African Law

Is the South African Law adequate?

South African law was not drafted with AI-generated deception in mind. The question is whether the current framework can still protect the integrity of criminal evidence, or whether Deepfakes expose gaps that require reform.

5.1 Strengths of the current framework.

South Africa does have a starting point for electronic evidence. Section 15 of the Electronic Communication and Transactions Act allows data messages to be admitted if their integrity can be shown, and courts have accepted that digital records must be authenticated before they are used.[27] The Criminal Procedure Act also permits experts evidence to explain technical material,[28] and our common law in S v Ndhlovu confirmed that electronic records are not excluded simply because they are not paper.[29] So, the law is not silent, it treats electronic evidence as real evidence, and it gives judge’s discretion to test realibility[30] That Flexibility is a strength when dealing with new technology.

5.2 Weakness Exposed by Deepfakes

The weakness is that “integrity” under section 15 of the ECT Act was meant for accidental alteration, not intentional AI fabrication.[31] The Act does not require a party who tenders a video to disclose whether it was AI- generated, nor does it create a presumption against synthetic media.[32] Practically, this means the onus falls on the opposing party to prove a recording is fake, often without access to forensic experts.[33] The SAPS forensic labs also limited capacity for Deepfakes’ detention, so the legal test is ahead of the technical capacity to apply it.[34] That creates inequality and delays.

5.3 Legislative Gaps

There is no South African statute that directly addresses Deepfakes in Criminal proceedings. The Cybercrimes Act 19 of 2020 criminalizes the unlawful creation and distribution of certain synthetic intimate images,[35] but it does not regulate the use of Deepfakes as evidence, nor does it set forensic or disclosure standards for courts.[36] The Law Reform Commisision has flagged AI and digital evidence as areas needing attention,[37] but no amendments to the ECT or CPA have been made yet.

5.4 International Comparisons

Other jurisdictions are moving faster. The United States has seen state laws requiring disclosure of AI- generated content in court, and proposals for digital watermarking of official recordings.[38] The European Unions’ AI Act takes a risk-based approach and imposes transparency duties on providers of generative AI systems,[39] which could support evidentiary rules later. The United Kingdom has updated judicial guidance to flag synthetic media as a specific authentication issue.[40] Compared to these, South Africa is still relying on general evidence principles without AI- specific rules.[41]

5.5 The Need for Reform

 The risk is twofold: courts may admit fabricated evidence and convict the innocent, or they reject genuine evidence because of “deepfake doubt”.[42] Both outcomes damage public trust and the right to a fair trial under section 35 of the Constitution.[43] Reform should there focus on the following things: 1. amending section 15 ECT Act to address deliberate AI manipulation, 2. issuing practice directives or guidelines on authentication of synthetic media, and 3. investing in forensic capacity and judicial training.[44] Without this, the law will remain adequate in theory but inadequate in practice.

Recommendations

If South African criminal courts are to cope with Deepfakes, the law, the institutions, and practice must change together. The aim is not to ban digital evidence, but to make sure courts can tell real from fake without undermining fair trial rights. Four linked reforms are needed.

6.1 Amend the Legislative Framework

Section 15 of the ECT Act should be updated to deal with deliberate AI manipulation, not only accidental corruption.[45] A new provision could require any party who tenders to a video or audio recording to disclose whether it was AI-generated or altered, and to provide provenance information where available.[46] Without this, the court is forced to assume authenticity until proven otherwise, which is risky in the age of deepfakes.[47]

6.2 Develop Court Rules and Practice Directives

Legislation alone is too slow for technology. The Rules Board and Chief Justice should issue practice directives on authenticating synthetic media.[48] These could set out a checklist for magistrates and judges: metadata review, chain of custody for digital files, and when to call a forensic expert.[49] The UK’s Crown Court guidance already flags synthetic media as a distinct authentication issue, and South Africa can adapt a similar model.[50]

6.3 Build Forensic and Judicial Capacity

A legal rule is useless if no one can apply it. SAPS Forensic Science Laboratories need access ro AI-based detection tools and training for analysts.[51] Legal Aid SA and the NPA must also have experts available so that poor accused persons are not disadvantaged compared to well-resourced parties.[52] At the same time, judicial officers should receive training on how Deepfakes work, how to assess expert reports, and how to avoid assessing expert reports, and how to avoid” deepfake doubt” that causes genuine evidence to be ignored.[53]

6.4 Promote Technical Safeguards and Cooperation

Courts cannot do this alone. Government should work with tech companies and SABS to encourage watermarking or provenance standards for official recordings, such as police body-cam or CCTV.[54] Internationally, the EU AI Act already imposes transparency duties on generative AI providers,[55] and SA can align with those standards. A national forensic database of known deepfakes artefacts would also help experts compare and verify exhibits faster.[56]

6.5 Protect Victims and Witnesses

Finally, reform must consider the human impact. The Cybercrimes Act already criminalizes some synthetic intimate images,[57] but prosecutors and victim support units need protocols for when deepfakes are used to intimidate witnesses or attack a complainant’s credibility.[58] Fast takedown processes and evidentiary protections will keep people willing to participate in criminal cases.

Conclusion

This article set out to answer whether South African criminal law is equipped to deal with deepfakes, and the findings show it is not yet. While the ECT Act and CPA provide a basic basis for admitting electronic evidence,[59] they were not designed for AI-generated fabrication and therefore leave courts exposed to both wrongful convictions and unwarranted” deepfake doubt”[60] The challenges identified are practical as much as legal authentication is harder, expert capacity is uneven, fair trial rights under section 35 of the Constitution are under strain, and victims can be intimidated by synthetic media[61] Compared with jurisdictions that have introduced disclosure rules, watermarking standards, and judicial guidance,[62] South Africa still relies on general principles without AI specific safeguards.[63] The law therefore needs targeted reform: amend section 15 ECT Act to address deliberate manipulation, issue of practice directives for authentication, and invest in forensic and judicial training.[64] Without these steps, public trust in digital evidence and in the criminal justice system itself will continue to erode as deepfakes technology becomes easier to use.

BIBLIOGRAPHY

Legislation: South Africa

Constitution of the Republic of South Africa 1996

Criminal Procedure Act 51 of 1977

Cybercrimes Act 19 of 2020

Electronic Communications and Transactions Act 25 of 2002

Legal Aid South Africa Act 39 of 2014

Magistrates’ Courts Act 32 of 1944

National Prosecuting Authority Act 32 if 1998

South African Judicial Education Institute Act 14 of 2008

Superior Courts Act 10 of 2013

Case Law

S v Ndhlovu 2002 2 SACR 325 (SCA)

Legislation: Foreign/ International

California, AB 730 2019

European Parliament and Council Regulation (EU) 2024/1689 on laying down harmonized rules on artificial intelligence OJ L 1689/1

Texas Business and Commerce Code 503.001(2024)

Books

Burchell J, Principles of Criminal Law (5th edn) (juta 2016)

Journal Articles

Chesney R and Citron D, DeepFakes: A Looming Challenge for privacy. Democracy, and National Security’ 107 Cal Rev 1753

Citron D and Chesney R, ‘Deepfakes and the New Disinformation War’ 1JFree Speech L299

Chesney H and Fleming S, ‘Deepfakes’; The Emerging Threat to Evidence’ Crim LR 34

Kemp G, Forensic Readiness in South African Criminal Justice’ 36 SACJ 512

Mason A, ‘Authenticating Digital Evidence in the Age of AI’ 138 SALJ 512

Schwartz M, ‘Jury Perception and Deepfake Evidence’ 18 Ohio st Tech LJ 44 (2019)(2021)(2023)(2022)

Official Publications and Reports

Judicial College. Crown Court Compendium Part 1 (2023)

Law Commission of South Africa, Discussion Paper on Cybercrimes and cybersecurity ( Project 126,2018)

Parliamentary Monitoring Group, ‘Cyber Crimes and Misuse of Technology: Briefing by SAPS’ (2022)

South African Law Reform Commission, Project 126 Issue Paper 34 (2021)

[1] R Chesney and D Citron, ‘Deepfakes: A Looming Challenge for Privacy, Democracy, and National Security’ (2019) 107 Cal L Rev 1753, 1755.

[2] H Chesney and S Fleming, ‘ Deepfakes: The Emerging Threat to Evidence’ Crim LR 34, 35.

[3] D Citron and R Chesney, ‘Deepfakes and the New Disinformation War’ (2019) 1 Journal of Free Speech Law 299, 301.

[4] Electronic Communications and Transactions Act 25 of 2002, s15; Criminal Procedure Act 51 of 1977, s212.

[5] H Chesney and S Fleming, ‘Deepfakes: The Emerging Threat to Evidence’ Crim LR 34,25.

[6] R Chesney and D Citron, ‘Deepfakes: A Looming Challenge for Privacy, Democracy and National Security’ (2019) 107 Cal L Rev 1753, 1756.

[7] D Citron and R Chesney, ‘ Deepfakes and the New Disinformation War’ (2019) 1 Journal of Free Speech Law 299, 302.

[8] M Schwartz, ‘Jury Perception and Deepfakes Evidence” (2022) 18 Ohio St Tech LJ 44,47.

[9] ibid 48.

[10] S Snail, Cyberlaw in South Africa 2nd ed (Juta 2012) 145.

[11] M Watney, ’Admissibility of Electronic Evidence in Criminal Proceedings: An Outline of the South African Legal Position’ (2009) 12 PELJ 46, 52.

[12] ibid 53.

[13] Cybercrimes Act 19 of 2020, Preamble: Snail (n1) 312.

[14] Electronic Communications and Transactions Act 25 of 2002, s15 (4); Law of Evidence.

[15] Electronic Communications and Transactions Act 25 of 2002, s15.

[16] H Chesney and S Fleming, ’Deepfakes: The Emerging Threat to Evidence’ Crim LR 34,36.

[17] A Mason, ’ Authenticating Digital Evidence in the Age of AI’ (2021) 138 SALJ 512, 518.

[18] Criminal Procedure Act 51 of 1977, s212.

[19] S V Ndhlovu 2002 SACR 325 (SCA) para 23.

[20] R Chesney and D Citron, ’ Deepfakes: A Looming Challenge for  Privacy, Democracy, and National Security” (2019) 107 Cal Rev 1753, 1772.

[21] D Kemp,’Forensic Readiness in South African Criminal Justice’ (2023)36 SACJ 201, 209.

[22] Consitutiona of the Republic of South Africa, 1996, s35(3)(i).

[23] M Schwartz,’ Jury Perception and Deepfakes Evidence’ (2022) 18 Ohio St Tech LJ 44, 51.

[24] D Citron and R Chensey. Deepfakes and the New Disinformation War’ (2019) 1 Journal of Free Speech Law 299,312.

[25] Parliamentary Monitoring Group,’ Cybercrimes and Misues of Technology: Briefing by SAPS’ (2022) 4.

[26] Law Commission of South Africa, Discussion Paper on Cybercrimes and Cybersecurity (Project 126, 2018) para 6.3.2. (2020).

[27] Electronic Communications and Transactions Act 25 of 2002’ s15.

[28]Criminal Procedure Act 51 of 1977, s212.

[29]S v Ndhlovu 2002 2 SACR 325 (SCA) para 23.

[30]J Burchell, Principle of Criminal La 5th ed (Juta 2016) 212.

[31]A Mason, ‘Authenticating Digital Evidence in the Age of AI ‘(2021) 138 SALJ 512, 517.

[32]H Chesney and S Fleming, ‘Deepfakes: The Emerging Threat to Evidence‘Crim LR 34,38.

[33]G Kemp, ‘Forensic Readiness in South African Criminal Justice ‘(2023) 36 SACJ 201, 208.

[34]Parliamentary Monitoring Group, ‘Cybercrimes and Misuse of Technology: Briefing by SAPS‘(2022) 6.

[35]Cybercimes Act 19 of 2020, s16.

[36]Law Commission of South Africa, Discussion Paper on Cybercrimes and Cybersecurity (Project 126,2018) para 6.4.1.

[37] ibid para 6.3.2.

[38]Texas Business and Commerce Code 503.001; California AB 730 2019.

[39]European Parliament and Council Regulation (EU) 2024/1689 AI Act, art 50.

[40]Judicial College, Crown Court Compendium Part 1(2023) 7-12.

[41]D Citron and R Chesney, ‘Deepfakes and the New Disinformation War‘(2019) 1 Journal of Free Speech Law 299, 310.

[42]M Schwartz ‘Jury Perception and Deepfake Evidence“(2022) 18 Ohio St Tech LJ 44, 53.

[43]Consttitution of the Republic of South Africa, 1996, s35 (3).

[44]South African Law Reform Commission, Project 126 Issue Paper 34 (2021) 44.

[45]Electronic Communications and Transactions Act 25 0f 2002, s 15.

[46]H Chesney and S Fleming, “Deepfakes: The Emerging Threat to Evidence ‘Crim LR 34, 39.

[47]A Mason, ‘Authenticating Digital Evidence in the Age of AI‘(2021) 138 SALJ 512, 520.

[48]Magistrates‘ Courts Act 32 of 1944; Superior Courts Act 10 of 2013.

[49]Judicial College, Crown Court Compendium Part 1 (2023) 7-12.

[50]ibid.

[51]G Kemp, ‘Forensic Readiness in South African Criminal Justice‘(2023) 36 SACJ 201, 210.

[52]Legal Aid South Africa Act 39 of 2014, s3; National Prosecuting Authority Act 32 of 1998, s22.

[53]South African Judicial Education Institute Act 14 of 2008, s4.

[54]European Parliament and Council Regulation (EU) 2024/1689 AI Act, art 50.

[55]ibid.

[56]R Chesney and D Citron, “Deepfakes: A Looming Challenge for Privacy, Democracy, National Security‘(2019) 107 Cal Rev 1753, 1780.

[57]Cybercimes Act 19 of 2020, s16.

[58]Law Commission of South Africa, Discussion Paper on  Cybercrimes and Cybersecurity (Project 126, 2018) para 6.5.3.

[59]Electronic Communications and Transactions Act 25 of 2002, s15; Criminal Procedure Act 51 of 1977, s212.

[60] A Mason, Authenticating Digital Evidence in the Age of AI‘(2021) 138 SALJ 512,520.

[61]Constitution of the Republic of South Africa, 1996, s35(3); Law Commission of South Africa, Discussion Paper on Cybercrimes and Cybersecurity (Project 126, 2018) para 6.3.2.

[62]European Parliament and Council Regulation (EU) 2024/1689 AI Act, art 50; Judicial College, Crown Court Compendium Part 1 (2023) 7-12.

[63] H Chesney and S Fleming. ‘Deepfakes: The Emerging Threat to Evidence‘Crim LR 34, 38.

[64]South African Law Reform Commission, Project 126 Issue Paper 34 (2021) 44.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top