Watch out for AI’s potential impact on legal privilege

Legal professional privilege, a long-held tenet of the legal system, is coming under scrutiny as artificial intelligence (AI) becomes more widely used in business and the legal profession.

By Vanessa Jacklin-Levin, partner at Bowmans

Some serious AI-related issues concerning legal privilege have arisen in courtrooms in South Africa and other jurisdictions, particularly the United Kingdom (UK) and the United States (US).

In South Africa, there are two main forms of legal professional privilege: legal advice privilege and litigation privilege.

Legal advice privilege protects client and legal adviser communications exchanged for the purpose of obtaining or giving legal advice. Litigation privilege protects communications between a client or its attorney and third parties, such as expert witnesses, made for the purpose of pending or contemplated litigation. In both cases, maintaining the confidentiality of the communications is an essential requirement for privilege: once it is lost or waived, by disclosure to other parties or publicly, the protection of privilege is lost. Another essential requirement for legal privilege is a qualified (natural person) legal adviser/attorney, without which there is no legal privilege.

Waiving or breaching any of the requirements of legal privilege can have dire consequences for a client’s legal position, as it may allow the opposing party in a dispute or litigation to access information that could damage the client’s case.

As AI use in legal matters becomes more widespread, we are seeing two trends that have the potential to result in privilege being lost or waived without the client realising it.

The perils of using AI for legal advice

Some companies have started using AI platforms for legal advice. They upload information onto an AI platform in the expectation that the output they receive is privileged.

It may not be.

In February 2026, Judge Jed S. Rakoff of the US District Court for the Southern District of New York issued a written opinion holding that AI-generated documents were not protected by privilege. The case involved a criminal defendant who had used the AI model Claude to research legal issues and draft a defence strategy. The defendant later shared this material with his defence lawyers and, when federal agents seized it, claimed privilege.

Judge Rakoff reasoned that Claude held no professional qualifications and owed no fiduciary or statutory duties, so it could not form an attorney-client relationship. Sharing the material with lawyers after the fact did not make it privileged.

Broadly, that accords with what we would expect in South Africa, where privilege attaches to communications with a qualified (natural person) legal adviser/attorney, not to research conducted independently by a client using AI.

Open or closed

In a business setting, the greatest AI-related risk to privilege arises when so-called legal advice is generated using a public or open AI platform.

It should be less risky to use a closed AI system, provided that the (natural) person making the request is legally qualified and acts in a legal capacity in the organisation. A closed system, used by a legally qualified person in a legal role, would not generally result in the loss of confidentiality and therefore of privilege. Even then, privilege will attach only where the in-house lawyer is giving legal, rather than commercial, advice.

However, such a system should be closed in a legal and contractual sense, not merely a technical one. This means the provider is contractually bound to keep inputs confidential, does not use them to train its models, and access is strictly controlled within the organisation.

Where an AI system is not subject to these constraints, it would be wise to exercise caution before using it to input confidential information and/or to generate legal advice.

Published to all the world

Another AI-related trend that is cause for concern is lawyers placing client-confidential information into public AI platforms. South Africa does not yet have definitive case law on this point, so we look to the US and UK for guidance.

Notably, many large public AI platforms have consumer terms stating that users should have no expectation of confidentiality or privacy, often because the information submitted may be used to train their large language models.

In October 2025, the UK judiciary published guidance that any information uploaded onto a public AI chatbot should be seen as published to all the world.

Six months later, in April 2026, in Munir v Secretary of State for the Home Department, the Upper Tribunal (Immigration and Asylum Chamber) held that a lawyer had waived privilege by uploading client letters onto a public AI tool.

The Tribunal also drew a clear distinction between public and closed AI systems, stating that closed enterprise environments do not present the same confidentiality risk.

Professional and data protection duties

For South African legal practitioners, the risk is not limited to privilege. Uploading client information to a public AI tool may also breach the duty of confidentiality under the Legal Practice Council’s Code of Conduct, and the security safeguards required of responsible parties under the Protection of Personal Information Act 4 of 2013 (POPIA), particularly where personal information is transferred to servers outside South Africa.

South African courts have also shown little tolerance for careless AI use. In Mavundla v MEC: Department of Co-operative Government and Traditional Affairs KwaZulu-Natal (2025) and Northbound Processing (Pty) Ltd v South African Diamond and Precious Metals Regulator (2025), the High Court referred legal practitioners to the Legal Practice Council after submissions cited non-existent, AI-generated authorities.

Practical steps

Organisations can reduce the risk by adopting a written AI use policy; restricting the use of public AI tools for confidential or legal matters; reviewing provider terms on confidentiality, data retention, model training and data location; routing legal queries through qualified legal advisers; and training staff on these risks.

These decisions underline the fundamental risks of AI in a legal context. AI tools can bring about significant efficiencies and opportunities, provided they are used appropriately and with proper governance and controls. If not, they can seriously undermine a client’s case.

Fortunately, courts and regulators, locally and in other jurisdictions, are beginning to provide guidance on the use of AI tools. Businesses and the legal profession should keep a close eye on these developments.

The post Watch out for AI’s potential impact on legal privilege appeared first on IT-Online.