News

Industries

Companies

Jobs

Events

People

Video

Audio

Galleries

Submit content

My Account

Advertise with us

Lessons from Heppner: Why you shouldn't use public AI to generate legal strategy

The convenience of modern AI tools is irresistible. These AI tools draft, summarise, brainstorm, and structure strategies in seconds. But in the recent case of United States v Heppner (Heppner), we are sharply reminded that convenience can come at a steep discovery cost.
Image source:
Image source: Freepik

If you build or refine legal strategies with a public AI tool, outside of a protected attorney–client setting, you may be laying a paper trail that is not privileged and is discoverable.

What happened in Heppner and why it matters

In this case, Heppner generated documents which he claimed were for legal strategy, using an AI assistant (Anthropic’s Claude) without consulting or being directed by a lawyer. When authorities later seized those documents, Heppner tried to shield them under attorney–client privilege.

The court rejected the claim because:

  1. the communications were not between client and attorney;
  2. the AI tool’s terms of service permitted certain disclosures, undermining confidentiality and any reasonable expectation of privacy;
  3. no legal advice from a lawyer was sought; and
  4. sharing the AI outputs with a lawyer afterwards did not retroactively create privilege.

The lesson to be learnt from Heppner is that, if you create legal strategy materials using an AI tool outside the attorney–client channel, do not expect privilege to save them.

The Heppner case also demonstrates that governance failures do not show up at the time that one purchases the AI tool but would likely appear when they are already in trouble.

Would a South African court treat this differently?

South Africa has not yet dealt with a case that touches on AI use and legal privilege. Nonetheless, if the Heppner case was before a South African court, it is unlikely that the matter would have been treated differently.

South African law protects two core forms of privilege:

  1. Legal advice (attorney–client) privilege – protects confidential communications between a client and a practising legal adviser made for the purpose of obtaining or giving legal advice.
  2. Litigation privilege – protects confidential communications or documents made for the dominant purpose of pending or reasonably contemplated litigation, usually within the attorney–client (and counsel) ecosystem or at their direction.

On those principles, Heppner’s materials drafted unilaterally with a third party AI tool, not at a lawyer’s behest, and subject to terms that dilute confidentiality, would almost certainly fall outside privilege in South Africa as well.

In other words, if the communication is not (i) with a practising attorney (or made at their direction for legal advice/litigation) and (ii) confidential, privilege will not attach.

That aligns with the thrust of the Heppner ruling.

The key South African takeaway is that for communications to enjoy privilege, they should be:
(a) with or at the direction of a practising attorney,
(b) for the purpose of obtaining legal or litigation advice, and
(c) kept confidential on systems and terms that preserve that confidentiality.

Finally, Heppner must not be taken as an AI panic story but should be considered as an AI governance story.

If you want the speed of generative tools and the shelter of privilege, design your workflows accordingly, route sensitive work through your legal practitioner and counsel, insist on confidentiality preserving environments, and memorialise purpose at creation.

Anything less is a discovery risk waiting to be exploited.

About Mandla Ngidi

Mandla Ngidi is an Associate at Adams & Adams. Mandla is a member of the SmartAIIP team at Adams & Adams focusing on everything where AI and IP collide.
More news
Let's do Biz