Artificial intelligence (AI) is no longer limited to asking a chatbot to draft an email or summarise a document. Legal AI platforms can now search case law, analyse large files, compare contracts, extract clauses, produce first drafts, and help firms organise knowledge that would previously have required hours of manual work. For a law firm, however, the question is not simply whether AI can save time, but a South African legal practitioner must also ask:
- Whether the tool is suitable for our legal system
- Where the information comes from
- What happens to client data
- Whether its sources can be verified
- Who remains responsible when the output is wrong?
That last question has become particularly important in South Africa. The Legal Practice Council’s 2026 Guide for ICT for Practice and Cyber Law expressly warns practitioners not to take AI outputs at face value and states that lawyers remain answerable for their research, arguments and representations even when AI is used.
This article, therefore, takes a practical look at the software currently available to legal practitioners, including South African options, the tasks different tools are designed to perform, recent cases showing what can go wrong, and a checklist firms can use before introducing AI into everyday practice.
Not all AI tools do the same job
One of the easiest mistakes a firm can make is treating every AI product as interchangeable. A general-purpose chatbot, a legal research database, and a contract-review platform may all use generative AI, but they are built for different tasks and carry different risks.
Before choosing a product, the first thing you must do is identify the problem the firm is actually trying to solve. Common legal workflows include:
- Legal research: Finding legislation, judgments and authorities relevant to a legal question.
- Drafting: Producing or refining first drafts of correspondence, pleadings, contracts and internal memoranda.
- Document analysis: Summarising bundles, comparing versions, extracting facts, and identifying clauses or issues across large document sets.
- Contract review: Spotting unusual clauses, comparing a document against a playbook and assisting with due diligence.
- Knowledge management: Searching the firm’s own precedents, opinions, templates and matter documents.
- Administrative work: Transcription, meeting summaries, task lists and routine internal communications.
A tool that performs one of these tasks well is not automatically reliable for another. The safest purchasing decision begins with the workflow, not with the product’s popularity.
AI applications currently used in legal practice
General-purpose generative AI
General-purpose tools remain attractive because they are easy to access and flexible, helping practitioners brainstorm, simplify complex material, create a structure for a document, summarise non-confidential text and improve the clarity of correspondence.
Their flexibility is also their limitation: a general chatbot should not be treated as a South African legal database, as it can generate plausible (very believable) but incorrect information or answers based on material that cannot be independently verified. For legal research, practitioners should therefore prefer tools that can point them back to authoritative source material.
Lexis+ with Protégé
LexisNexis South Africa now offers Lexis+ with Protégé, combining generative AI functionality with LexisNexis legal content. Its current offering includes legal research, drafting, and workflow functions and is designed to work with both trusted LexisNexis sources and organisational knowledge.
CoCounsel and Westlaw
Thomson Reuters CoCounsel is designed for professional legal workflows and integrates AI with Thomson Reuters legal and practical content. Depending on the product and jurisdiction, functions include research, document review, summarisation, drafting and analysis.
For South African firms, jurisdictional coverage should be checked before subscribing. A sophisticated international product is not necessarily the right research tool if the firm’s work depends primarily on South African authorities.
Harvey and other enterprise legal AI platforms
Harvey has become a prominent enterprise legal AI platform used by major law firms and legal departments, designed for workflows such as research, due diligence, drafting and large-scale document analysis. In September 2026, Reuters reported another major funding round for Harvey, illustrating the rapid investment taking place in legal AI.
Platforms of this kind are particularly relevant to larger firms that want AI integrated into multiple workflows. Smaller firms should not assume that an enterprise platform is automatically the best choice, because (1) implementation, (2) security, (3) jurisdictional coverage, (4) training, and (5) cost are all equally important.
Contract-focused tools
Transactional practices may also consider specialised products such as Spellbook, which works within Microsoft Word to assist with contract drafting and review. Other contract-focused systems can compare clauses against a firm’s preferred language, identify unusual provisions and accelerate first-pass review. These tools can reduce repetitive work, but they do not know the client’s commercial priorities unless those priorities are properly supplied and reviewed. This is why it is important that the practitioner still decides what rights, obligations and risks are appropriate for the particular transaction.
South African legal AI: LexCore
South African firms do not have to rely solely on international platforms, as LexCore by LexVivens is a locally developed AI legal research and drafting platform built specifically for South African legal professionals.
LexCore offers natural-language legal research, precedent matching, document analysis, and drafting functions. According to LexVivens, its case-law research environment is grounded in verified South African source material and links judgments back to SAFLII. The platform also supports multiple file types, including PDFs, Word documents, screenshots, audio and handwritten material.
This local grounding is important because a South African practitioner generally needs more than a fluent answer; they require the correct jurisdiction, current legislation, genuine cases, and a route back to the underlying authority. At the same time, firms should apply the same due diligence standard to every vendor. Marketing claims about accuracy, security, or the absence of hallucinations should be tested against the firm’s own requirements, since it remains a lawyer’s professional duty to verify important legal propositions, and this duty does not disappear because a platform is purpose-built for law.
Other options worth exploring
The legal AI market is expanding quickly, and firms may encounter many products beyond the best-known names. Rather than creating a long list that will quickly become outdated, it is more useful to group the alternatives by purpose:
Research platforms: Look for products that identify their legal corpus, provide citations, and allow the practitioner to open the underlying authority.
Document-review tools: Useful for due diligence, discovery, and large-matter files where the firm needs to classify, compare, or extract information.
Drafting assistants: Useful for first drafts, clause suggestions, and document refinement, provided the final document is reviewed by the responsible practitioner.
Firm-specific knowledge tools: Useful where a practice wants to search its own precedents and institutional knowledge rather than send documents into a public system.
Transcription and meeting tools: Potentially useful for consultations and meetings, but only after considering confidentiality, consent, storage and data-processing implications.
A firm’s technology stack may therefore contain several AI tools rather than one product that does everything. The important issue is to define what each tool may be used for and where human review is compulsory.
What happens when lawyers rely on AI without checking it?
South African courts have already provided clear warnings. The recurring problem is not merely that AI can make mistakes; it is that practitioners have placed AI-generated material into professional work without independently verifying it.
Mavundla: fabricated authorities and a warning from the High Court
In Mavundla v MEC: Department of Co-Operative Government and Traditional Affairs KwaZulu-Natal and Others, the legal team relied on authorities that could not be found. It emerged that the cases had been sourced from ChatGPT and had not been independently reviewed. The High Court described the conduct as careless and emphasised that AI-generated legal research still requires independent reading and verification. The matter became an important warning from South Africa that efficiency cannot replace professional judgment.
Northbound Processing: even a legal AI tool can hallucinate
The lesson became even clearer in Northbound Processing (Pty) Ltd v South African Diamond and Precious Metals Regulator and Others. The High Court discovered non-existent authorities in the heads of argument. Counsel confirmed that the incorrect citations appeared to be hallucinations produced by a legal AI tool called Legal Genius. The court accepted that there had not been a deliberate attempt to mislead it, but stressed that negligence can still have serious professional consequences. The issues were referred to the Legal Practice Council for investigation.
This case is particularly important for firms choosing software, as it clearly shows that a product marketed as a legal AI tool does not remove the practitioner’s obligation to verify its output.
The risk is continuing internationally
The problem is not confined to South Africa. In September 2026, Reuters reported that the New Mexico Supreme Court held a lawyer in contempt and imposed a US$5,000 fine after a brief containing fabricated police testimony and fictitious witnesses generated by ChatGPT. The details differ from case to case, but the professional lesson is consistent: the practitioner who signs, files, or relies on the work remains responsible for it.
An ethical AI checklist for law firms
The LPC’s 2026 ICT and Cyber Law guide makes the underlying principle clear: AI is not a replacement for human responsibility and oversight. A practical firm policy can turn that principle into everyday safeguards.
Before adopting a tool
- Decide exactly which tasks the tool is permitted to perform.
- Confirm whether the platform actually contains the South African law required for the firm’s work.
- Ask where legal answers come from and whether users can open and verify the underlying authority.
- Understand where prompts and documents are stored, who can access them, and whether they may be used to train models.
- Use non-confidential sample matters to assess accuracy, usability and failure points.
- Not every staff member necessarily needs the same permissions or the ability to upload sensitive matter information.
When using AI on a matter
- Do not place privileged, confidential or personal information into an AI system unless the firm’s use of that system is authorised and its data protections have been assessed.
- Open the actual case, legislation or regulation and confirm that it exists, is current and supports the proposition for which it is cited.
- A genuine case can still be accompanied by a fabricated quotation or paragraph reference.
- Treat AI-generated drafting as a first draft, not as finished professional work.
- The tool can assist with analysis; it cannot assume the practitioner’s professional responsibility to the client or court.
- Check court directions, client arrangements, and applicable professional requirements to determine whether the use of AI should be disclosed.
- Candidate attorneys, paralegals, and other team members should understand the firm’s approved tools and verification requirements.
What firms should avoid
- Do not use a chatbot as the sole source of legal authority.
- Do not copy AI-generated citations into court documents without opening the sources.
- Do not upload client files simply because a tool allows it.
- Do not assume a “legal AI” label guarantees accuracy.
- Do not allow efficiency to become fictional billing.
- Do not deploy AI without a firm policy.
A five-question test before relying on AI output
- Can I identify where this information came from?
- Have I independently checked every legal authority and quotation?
- Would I be comfortable explaining to the client or court how this work was produced?
- Have I protected confidential and personal information?
- Am I personally satisfied that the final work is legally correct and appropriate for this matter?
If the answer to any of these questions is no, the output is not ready for professional use.
Resources for South African practitioners
Legal Practice Council: Guide for ICT for Practice and Associated Aspects of Cyber Law – current LPC training guidance covering AI, hallucinations, verification, cybersecurity and professional responsibility.
American Bar Association Formal Opinion 512 – useful comparative guidance on competence, confidentiality, communication, supervision, candour and fees when lawyers use generative AI.
Practitioners should also monitor court practice directives and LPC communications. AI products and professional expectations are developing quickly, so a firm’s policy should be reviewed rather than treated as a once-off exercise.
Conclusion
The most useful AI tool for a law firm is not necessarily the one with the most features, but rather the one that solves a defined problem, fits the firm’s jurisdiction and workflow, protects client information, and allows the practitioner to verify its output. AI can shorten the mechanical parts of legal work, but it cannot transfer professional responsibility from the lawyer to the software.
South African courts have already made that boundary clear. Firms that combine appropriate technology with verification, supervision and independent legal judgment are better placed to gain the efficiencies of AI without compromising the standards on which legal practice depends.
About the author
Theshaya Naidoo is a PhD (Law) Candidate and Canon Collins Scholar. Her research focuses on 'The Legal & Ethical Implications of Neurotechnology on the South African Criminal Justice System'. She holds an LLM in Medical Law, where her thesis focused on 'The necessity of sui generis AI regulation in South Africa'. Theshaya attended Gawie le Roux Institute of Law’s Attorneys’ admission exam courses and law school.
Last updated on 14 October 2026.