Enforcement gaps in South Africa’s gender-based violence laws after the 2021 amendments

Introduction

South Africa does not lack gender-based violence law. The Domestic Violence Act 116 of 1998 has been around for more than twenty-five years, the Criminal Law (Sexual Offences and Related Matters) Amendment Act 32 of 2007 rewrote what counts as rape, and in January 2022 the President assented to three more statutes, among them the Domestic Violence Amendment Act 14 of 2021 and the Criminal and Related Matters Amendment Act 12 of 2021. Most of the domestic violence changes only took effect on 14 April 2023, so the newest rules are still young. Even so, the lawyers writing about them keep circling the same worry, which is that the text keeps improving while the place where a woman actually meets the state stays pretty much as it was (Sibisi, 2023), a worry Nortje and Hull (2024) share.

This paper argues that the 2021 amendments tighten the statute where drafting is easiest to fix, while the evidence points to weaker links earlier in the chain, namely how police record and investigate a complaint (Machisa et al., 2023) and how a protection order application is processed at the court counter (Backe et al., 2025). If that is right, new definitions and stricter bail rules will reach only a share of cases, welcome as they are in their own right. The paper also tests the argument against the late 2025 classification of gender-based violence and femicide as a national disaster, which turns out to be narrower than the headlines suggested. It moves through what the amendments changed, where cases leak out of the system, how protection orders work in practice, what bail and judicial development can and cannot do, and the disaster classification, before taking a position in the conclusion.

What the 2021 amendments actually changed

The Domestic Violence Amendment Act added coercive and controlling behaviour to the definition of domestic violence (Botha, 2025). It also lifted the penalty for breaching a protection order to as much as five years for a first offence and ten for a repeat one, and it removed the references to imminent harm that used to guide officers deciding whether to arrest (Sibisi, 2023). The same article adds that victims can now apply for orders online, and that a new section 4A creates a safety monitoring notice that obliges police to check on a victim who shares a home with the abuser. On the bail side, Nortje and Hull (2024) explain that a person accused of a domestic violence offence now has to ask a court for bail, because police and prosecutors can no longer release the accused before a first appearance. That is a real change. Nobody should wave it away.

What stands out is how little these authors agree about where the trouble lies. Sibisi (2023) worries about drafting, for instance emergency monetary relief with no stated time limit, and about officers losing a defined test for arrest once imminent harm disappeared, which he thinks invites unlawful arrests and civil claims. Singo (2023) goes after the new definitions themselves, calling coercive and controlling behaviour vague, overbroad and ambiguous, although Singo suggests a constitutional challenge is unlikely to succeed given how much the concept matters. Botha (2025) is more pessimistic about people than about paper, doubting that an understaffed police service will make the monitoring notices work well. So there are three pressure points: drafting, constitutionality and capacity. Parliament can fix the first two in a sitting or two. The third is a different animal altogether, and this paper’s view is that it is the one that decides outcomes.

Where cases leak out before the statute can help

The best national evidence on where things go wrong comes from Machisa et al. (2023), who followed 3,952 rape cases reported to the police in 2012. Police closed about 35 per cent without sending them to a prosecutor, prosecutors declined 31 per cent, another 16 per cent were enrolled and later struck off the roll, and only 9 per cent ended in a conviction for a sexual offence. So the biggest single drop happened before a prosecutor ever saw the docket. These cases were handled under the 2007 Act and predate the 2021 amendments by almost a decade, and rape dockets are not the same thing as domestic violence dockets, so the figures cannot show whether anything has moved since. It would be unwise to stretch them. Still, Nortje and Hull (2024) describe distrust of the police as one reason so many victims never report in the first place, which points the same way.

The detail inside the dockets is actually more telling than the headline percentages. About a third of the medical examination forms were missing, 348 selected dockets could not be found or were too incomplete to use, and cases where a forensic sample had reached the laboratory had lower odds of being declined by prosecutors (odds ratio 0.57). A scene visit by the investigator was linked to higher odds of a guilty verdict. None of this is exotic. It is basic police work, and it is what the authors have in mind when they call for thorough investigation and continuing training on rape stereotypes (Machisa et al., 2023). Gumani (2022) looked at the same problem from the officers’ side, interviewing fifteen members of the Visible Policing and Detective Service units in the rural Vhembe district of Limpopo. The study reports that cultural customs and household dynamics shaped how officers responded, producing what it calls incident-focused and evasive strategies. Small sample, one district, but the finding fits.

Courts have started treating this as a legal wrong in its own right. In AK v Minister of Police, decided by the Constitutional Court in April 2022, the police were found negligent in the search for an abducted woman: no basic foot search at first, a dog handler who stopped short of an area he knew about, CCTV footage reviewed late, and DNA evidence followed up slowly. The Court measured the service against what a reasonable organ of state would have done and described the duty as one of means rather than result. Mukheibir (2025) reads the case as one more step in a line running back to Carmichele, and then asks the uncomfortable question of why so little has changed on the ground after two decades of damages awards, noting that the money comes out of the same public purse that pays for policing. Botha (2025) adds that police are severely understaffed and often fail to act on existing protection orders. Put those together with the attrition figures and damages litigation starts to look like a late, individual and expensive remedy. It sets a standard after the harm is done, and whether it lifts performance is another matter.

The protection order counter

Protection orders are the everyday tool of the Domestic Violence Act, and Backe et al. (2025) show how unevenly the tool is handed out. Drawing on fieldwork, interviews and focus groups gathered between 1999 and 2022 in Western Cape magistrates’ courts, they found that application forms differ from court to court even though prescribed forms exist. One carried a dedicated urgency section while others had none, and one listed twenty subtypes of domestic violence where others listed fewer. Clerks hold a lot of discretion and sometimes act as gatekeepers, and interim orders, return dates and service vary with the magistrate and the court. Some courts collected orders daily and others twice a week, which can stretch the wait for protection. Sibisi (2023) had said the Act was drafted with more care than its implementation received, and these courts look like a working illustration.

The counter-level findings go further. Police were reported to refuse to act while an application was pending and to respond more readily to visible injury than to emotional or financial abuse, and warrants of arrest were not routinely issued with interim orders, so some applicants had to return to court before police would move (Backe et al., 2025). Applicants were also blamed for abusing the process or failing to follow it, even though the process itself was inconsistent. This matters for the new definition of coercive and controlling behaviour, because that is the kind of abuse that leaves no bruise for a front-desk officer to see (Botha, 2025). Backe et al. conclude that the shortcomings stem largely from inconsistent implementation and not from the Act itself, and that the 2021 amendments, which do give guidance on return dates where parties are absent, are unlikely to reach the procedural differences hidden between courts.

Two cautions are in order. The Western Cape research rests on ten courts, and some of the fieldwork is more than twenty years old, so the picture may be harsher than today’s. What it describes does not map neatly onto rural Limpopo either, where Gumani (2022) found officers whose responses were shaped by local custom as much as by any form. Still, the two studies disagree less than they seem to. Both place the decisive moment with an individual official, a clerk or a constable, using discretion that the statute never mentioned. If outcomes really get decided there, then Singo’s (2023) worries about the wording of a definition are real but they sit a long way upstream, and this whole thing looks more like an administration problem than a drafting one.

Bail, the courts and the limits of reform

Bail is where Parliament has done the most visible work, and Nortje and Hull (2024) still think it has not gone far enough. They propose a reverse onus in serious cases where the accused has confessed, was caught in the act or has earlier convictions for gender-based violence, tighter conditions with harsher consequences for breach, automatic High Court review of bail conditions in serious cases, and refresher training for magistrates. Their article is doctrinal rather than empirical, so the proposals rest on argument and on cases such as Carmichele and not on data about how bail is currently decided. That is a weakness, though a forgivable one given how thin the data are. A second look at the problem comes from Machisa et al. (2023): police arrested or charged the accused in just under 58 per cent of the rape cases, which means bail rules, however strict, are only ever tested in a portion of them.

Existing rules can do more than they are given credit for. In NH v S the Johannesburg High Court dismissed an appeal against refusal of bail pending a petition by a man convicted of rape, kidnapping and breaching section 17 of the Domestic Violence Act. Because rape is a Schedule 6 offence, he had to show exceptional circumstances, and a generic affidavit did not do it. The court also pointed to the risk to a complainant who had already seen an interdict ignored, and breach of an order is, after all, a criminal offence under section 17, as Sibisi (2023) sets out. A careful magistrate and a prepared prosecutor can apparently reach this result under the old test, which fits Nortje and Hull’s (2024) call for refresher training better than it fits a call for yet more legislation. One decision proves little, and nothing in the sources shows how typical it is.

The courts shape the reach of these laws in other ways too. Mokone (2021) analyses Tshabalala v S; Ntuli v S, where the Constitutional Court held that common purpose can apply to common-law rape, so that men who joined in a 1998 attack on eight women in Tembisa need not each have personally penetrated a victim to be convicted. Mokone treats that as the judiciary using section 39(2) of the Constitution to develop the law, and adds that since the laws already exist, the remaining problem lies in social attitudes. Stal (2023) takes the Coko decision, in which the Eastern Cape High Court overturned the rape conviction of a 23-year-old paramedic, and argues that it is consistent with the principle that intention is tested subjectively. Stal reports that critics read the decision as treating force as necessary for rape, and argues that those critiques partly misread it. Whichever reading is right, the analysis suggests that proof of intention stays a pressure point in contested-consent cases, and no 2021 amendment touches it.

Botha (2025) makes a related point from the opposite direction, about abused women who kill. The article argues that the quickest fix is not a new statute at all but a Supreme Court of Appeal decision restoring the pre-Eadie test for non-pathological criminal incapacity, and it notes that no reported South African case has yet succeeded with private defence for an abused woman who killed in a non-confrontational setting. Whatever one makes of the proposal, the lesson carries across to bail and consent. Courts, not Parliament, decide a good deal of what gender-based violence law means in practice, a point Mokone (2021) also draws from Tshabalala.

What the national disaster classification adds

It helps to be exact about what happened in November 2025, since the status is narrower than the phrase suggests. On 21 November 2025 the head of the National Disaster Management Centre classified gender-based violence and femicide as a national disaster under section 23 of the Disaster Management Act 57 of 2002 (South African Government News Agency, 2025). The decision came from the Centre’s head rather than from legislation, it followed a reassessment of earlier reports and new submissions, and the department said it does not invoke emergency powers and works through existing legislation and contingency arrangements. In practical terms, then, the classification is a coordination step. The related institutional step is the National Council on Gender-Based Violence and Femicide Act, signed on 24 May 2024, which Mukheibir (2025) treats as positive provided it is actually implemented.

Does that change the argument? Slightly, and mostly as an opportunity. A classification built on existing structures could in principle be used to push at the front end, through audits of docket completeness of the kind Machisa et al. (2023) describe, a single prescribed set of protection order forms actually used in every court, which Backe et al. (2025) show is not yet the case, and magistrate training of the kind Nortje and Hull (2024) propose. None of those needs a new statute, and none needs emergency powers. Whether they happen is another thing, and I could find no published evaluation of the classification’s effect, which is hardly surprising so soon. The risk is that it becomes one more announcement sitting above the same counters.

Conclusion

The 2021 amendments are worth having. Coercive and controlling behaviour is now named, bail for domestic violence accused has moved into the courtroom, and penalties for breaching protection orders have gone up (Sibisi, 2023; Botha, 2025). Nothing in this paper is meant to dismiss that work, and Singo’s (2023) drafting concerns can be taken up by Parliament whenever it chooses.

But the evidence gathered here points to a gap that the statutes were never designed to close. Cases drop out at the police stage (Machisa et al., 2023), officers respond to local custom and visible injury (Gumani, 2022), and the protection order a woman gets depends partly on the court she happens to walk into (Backe et al., 2025). Damages claims and new definitions both arrive after those moments, not before them.

The position taken here is that Parliament has reached diminishing returns on drafting. The next gains probably lie in dull things like complete dockets, one court form, bail decisions made by officials trained to use the test already in place, and police staffed well enough to act on an order the day it is served. Mukheibir (2025) is blunt that courtroom victories have not yet turned into lower violence, and the national disaster classification will likely be judged the same way. The sources leave real gaps, mainly that the strongest attrition data predate the amendments and the court-counter research comes from one province (Machisa et al., 2023; Backe et al., 2025), so this is a best reading of thin evidence. Still, nothing in what was reviewed suggests that a tighter statute, on its own, will get a complaint any further than the front desk.

Reference list

Backe, E.L., Moult, K. and Waldman, J. (2025) Dissonance, discordance, and disparity: the impact of differences in papers, processes, and people on justice for domestic violence survivors in South Africa. International Journal for Crime, Justice and Social Democracy, 14(3), pp. 80-90. doi:https://doi.org/10.5204/ijcjsd.3164

Botha, R. (2025) Criminal justice for female victims of domestic violence: time to deal defencelessness a final blow. Obiter, 46(4), pp. 968-985. doi:https://doi.org/10.17159/pjdnzz54

Gumani, M.A. (2022) Police interventions in victims’ encounters of intimate partner violence in a rural setting. South African Journal of Psychology, 52(3), pp. 379-391. doi:https://doi.org/10.1177/00812463221086379

Machisa, M., Jina, R., Labuschagne, G., Vetten, L., Loots, L. and Jewkes, R. (2023) Factors associated with rape case attrition in the South African criminal justice system: a national cross-sectional study. The British Journal of Criminology, 63(3), pp. 588-614. doi:https://doi.org/10.1093/bjc/azac044

Mokone, G. (2021) The constitutional role of the judiciary in cases of sexual GBV: an analysis of Tshabalala v S; Ntuli v S 2020 (5) SA 1 (CC). Obiter, 42(2), pp. 406-420. doi:https://doi.org/10.17159/obiter.v42i2.11928

Mukheibir, A. (2025) The law of delict and gender-based violence in South Africa. Obiter, 46(4), pp. 950-967. doi:https://doi.org/10.17159/y5w1aj39

Nortje, W. and Hull, S. (2024) Uprooting a culture of gender-based violence in South Africa: critical approaches to bail, policing and awareness. Obiter, 45(2), pp. 295-309. doi:https://doi.org/10.17159/obiter.v45i2.16254

Sibisi, S. (2023) The Domestic Violence Act 116 of 1998: offences, defences, economic abuse, imminent harm and the crime of domestic abuse: what is new? Obiter, 44(2), pp. 331-347. doi:https://doi.org/10.17159/obiter.v44i2.14322

Singo, D. (2023) Coercive and controlling behaviour in the Domestic Violence Act. South African Law Journal, 140(4), pp. 763-794. doi:https://doi.org/10.47348/SALJ/v140/i4a4

Stal, S.J. (2023) Does mistaken belief in consent constitute a defence in South African rape cases? Potchefstroom Electronic Law Journal, 26, pp. 1-20. doi:https://doi.org/10.17159/1727-3781/2023/v26i0a15002

Legislation and case law

AK v Minister of Police [2022] ZACC 14 (CCT 94/20), Constitutional Court of South Africa, 5 April 2022. Available at: https://www.saflii.org/za/cases/ZACC/2022/14.html (Accessed: 18 October 2026).

Criminal and Related Matters Amendment Act 12 of 2021. Available at: https://lawlibrary.org.za/akn/za/act/2021/12/eng@2022-01-28 (Accessed: 18 October 2026).

Criminal Law (Sexual Offences and Related Matters) Amendment Act 32 of 2007. Available at: https://www.lawlibrary.org.za/akn/za/act/2007/32 (Accessed: 18 October 2026).

Domestic Violence Act 116 of 1998. Available at: https://lawlibrary.org.za/akn/za/act/1998/116 (Accessed: 18 October 2026).

Domestic Violence Amendment Act 14 of 2021. Available at: https://lawlibrary.org.za/akn/za/act/2021/14 (Accessed: 18 October 2026).

NH v S (A128/2022) [2022] ZAGPJHC 1031 (21 November 2022), High Court of South Africa, Gauteng Local Division, Johannesburg. Available at: https://www.saflii.org/za/cases/ZAGPJHC/2022/1031.html (Accessed: 18 October 2026).

South African Government News Agency (2025) CoGTA welcomes classification of GBVF as a national disaster. Available at: https://www.sanews.gov.za/south-africa/cogta-welcomes-classification-gbvf-national-disaster (Accessed: 18 October 2026).