In January 2025, the KwaZulu-Natal High Court examined heads of argument in a matter before it. Nine authorities were cited.
Seven did not exist.
They had been generated by ChatGPT. The court described the conduct as "irresponsible and unprofessional." The matter was referred to the Legal Practice Council for investigation.
This was not a first. Parker v Forsyth in 2023 had already seen fabricated AI-generated case law submitted to a South African court. It was not the last either — Northbound Processing (Pty) Ltd v The South African Diamond and Precious Metals Regulator followed with costs orders and another mandatory LPC referral.
South African courts have made their position clear. They are not interested in excuses about time pressure or unfamiliarity with AI. If you submit it, you own it. The fact that a machine wrote it is not a defence. It is an aggravating factor.
This is the landscape you are practising in. And it is changing faster than most attorneys are acknowledging.
What Is Actually Happening in the SA Legal Market Right Now
While many attorneys are still deciding whether to "try AI," the largest firms in the country have already moved.
Bowmans implemented Harvey AI firm-wide. ENS followed on 1 May 2025, using it to automate document review and legal research. Herbert Smith Freehills Kramer began piloting Harvey AI across selected practice areas shortly after their merger. These are not small experiments — these are firm-wide rollouts at South Africa's most prominent legal practices.
The message this sends to the rest of the profession is not subtle: AI is no longer a curiosity. It is operational infrastructure at the top end of the market.
The numbers are stark. South Africa's legal technology market is growing at a fast rate, projected to nearly double by 2030. According to Thomson Reuters' 2024 Future of Professionals report, AI has the potential to automate up to 23% of a lawyer's working hours and unlock the equivalent of R1.9 million in potential new billable value per lawyer, per year, if that recaptured time is redirected to higher-value work.
That is the upside. But the downside is real, documented, and growing in South African courts.
The Two Failure Modes Destroying SA Attorneys Right Now
Failure Mode 1: Avoidance
Some attorneys have decided AI is too risky, too complicated, or too foreign to their practice. They are not entirely wrong — the risks are real. But they are making a strategic error: the cost of doing nothing is not zero. It is compounding.
Clients are already beginning to ask why their attorneys cannot turn work around faster. In five years, AI-augmented efficiency will be the baseline expectation, not a premium offering. The Legal Practice Council is expected to integrate AI competence into CPD requirements by 2026. This means that within months, demonstrating AI fluency will no longer be optional — it will be a professional obligation.
Failure Mode 2: Blind Trust
This is the failure mode that ends careers. It is the attorney who uses ChatGPT to research case law, assumes the output is correct because it sounds authoritative, and submits it without verification. It is the attorney in Mavundla. It is the attorney whose matter is now on the LPC's desk.
Generative AI does not know when it is wrong. It will happily generate a beautifully formatted legal citation that has never existed in the history of South African law. That is not a flaw that will be patched. It is an inherent characteristic of how these models work.
In Mavundla, the court reportedly tested ChatGPT itself, asking it to confirm the cited cases. It did — all of them, including the seven that did not exist. Using AI without understanding this is not a time-saving strategy. It is a malpractice risk.
The South African-Specific Issues You Cannot Ignore
POPIA Compliance
The Protection of Personal Information Act creates obligations that extend directly to how you use AI tools in client matters. POPIA treats client information as protected personal data, and transferring it to external processors — especially outside South Africa — requires specific safeguards. Most attorneys using these tools have not conducted the vendor due diligence that POPIA requires.
SA Case Law Gaps in Global AI Tools
Tools like ChatGPT and even Harvey AI are primarily trained on international — predominantly American — legal materials. Their familiarity with South African legislation, LPC rules, and SA case law is limited and inconsistent. This is precisely why Mavundla happened. When using any AI tool for SA-specific legal research, you cannot assume the model has adequate training on local materials. Every citation must be verified against authoritative local sources: SAFLII, Jutastat, LexisNexis South Africa, or the official Government Gazette.
The LPC's Increasing Scrutiny
South African courts have now established a clear pattern: hallucinated AI citations result in mandatory LPC referrals. The LPC itself is moving to make AI competence a formal CPD requirement. This is not a distant regulatory concern. It is a present reality that has already affected the careers of practitioners who were not paying attention.
What Responsible AI Use Actually Looks Like in SA Practice
- Verify every citation against SAFLII or Jutastat. Every time. This is non-negotiable. No AI-generated case citation should enter any court document without being independently verified against a primary authoritative source. Build this into your workflow as a mandatory step, not an optional check. If you cannot find the case on SAFLII or in Jutastat, the case does not exist for your purposes.
- Treat AI output as a first draft from a very confident, very unreliable junior. AI tools produce excellent first drafts of contracts, pleadings, and legal research memos. They are genuinely useful starting points. But they require the same substantive review you would give a candidate attorney's first attempt — because that is essentially what they are.
- Do not upload client documents to consumer AI tools. Use tools that offer enterprise-grade data privacy, explicit data processing agreements, and compliance with POPIA's requirements for cross-border data transfers. When in doubt, consult your POPIA compliance framework before onboarding any new AI tool for client work.
- Document your AI use in your file. In a post-Mavundla environment, a well-documented AI supervision protocol is your professional insurance. Note what tools you used, for what purpose, and what verification steps you took. If your work is ever challenged, this paper trail demonstrates that you treated AI as a tool that requires supervision — not as an authority.
- Consider SA-built tools for SA law. Adjunct AI was built specifically for South African legal research, with SA legislation and case law at its core. For SA-specific work, a purpose-built local tool may outperform a globally-trained model with limited SA coverage.
The Bigger Picture
The attorneys who thrive in the next decade will not be the ones with the most years of experience or the most prestigious pedigrees. They will be the ones who combine rigorous legal judgment with genuine AI fluency — who understand both the law and the tools well enough to use each appropriately, who can deliver more to clients in less time, at a competitive cost, without creating malpractice exposure.
That combination is still rare. Which means the window to build a genuine competitive advantage is still open.
The Mavundla matter will not be the last of its kind. The LPC's scrutiny will intensify, not relax. POPIA enforcement is maturing. And the firms that have invested in Harvey AI are already pulling ahead.
The question for every South African attorney is the same one it always is: are you going to be ahead of this, or behind it?
This blog exists to help you get — and stay — ahead.
📚 Cases Referenced in This Article
Nothing in this article constitutes legal advice. Always consult a qualified attorney for your specific matter.