Regulation Tracker
AI rules for legal practice in South Africa
As of 2026-09-04, 4 official documents govern AI use by lawyers, judges or litigants in South Africa. 1 requires disclosure or certification of AI use, 2 impose an explicit verification duty, and 1 prohibits some use outright. The incident tracker records 5 incidents in this jurisdiction.
South Africa category map →
SA Judiciary Conference
Resolutions adopted by the South African Judiciary on 16 July 2026. Section 5 directs the Judiciary Administration and Information Technology Steering Committee to finalise the draft judiciary AI policy, requires the Judiciary to retain governance over AI procurement and deployment, restricts initial AI use to efficiency tools, and mandates ongoing judicial education on responsible AI use.
Policy · Supervision · Competence · Record-keeping · Verification duty
2026-07-16In force
LSSA AI Committee
Revised draft ethics guidelines on generative AI for South African legal practitioners, published 1 July 2026 in De Rebus by drafters working with the LSSA AI Committee after a 2026 consultation. Practitioners are fully responsible for AI-assisted work, must assess local relevance and bias, and must address cybersecurity risks. Not formally adopted by the LSSA or Legal Practice Council.
Guidance · Verification duty · Competence · Confidentiality
2026-07-01Proposed
DCDT
South Africa's draft national AI policy, gazetted for comment on 10 April 2026, proposed new oversight bodies including a National AI Commission. It was withdrawn in its entirety on 12 June 2026 after reporting established that several of its 67 academic citations — journals and authors alike — did not exist, apparently AI hallucinations in the policy's own drafting.
Policy · Risk classification · Disclosure
2026-04-10Withdrawn
OCJ
A draft policy circulated for comment by the Office of the Chief Justice in April 2026 setting binding standards for responsible, ethical, secure and transparent AI use in the South African Judiciary. It distinguishes adjudicative from administrative functions, prohibits AI from adjudicative work, and imposes an explainability requirement. Not adopted as of 4 September 2026.
Policy · Prohibition · Supervision · Competence · Record-keeping
2026-04-01Proposed
Descriptive reference only: this page records what each rule requires. Missing a document? Tell us.