South Africa Confirms Revised AI Policy Has Real Citations; AI Wrote the Papers to Make It So

PRETORIA—The South African Department of Communications and Digital Technologies announced Tuesday that its revised National Artificial Intelligence Policy has cleared a full citation audit, with all 67 references confirmed accurate—a significant improvement over the first draft, which the government withdrew in January after at least six of its citations were determined to be entirely fabricated by the AI system commissioned to write the document.

"Every source in this version is real," said a departmental spokesperson. "We are confident in that."

A subsequent independent audit found that 23 of the 67 citations—among them a peer-reviewed paper on ethical AI deployment, a working paper from an OECD subdirectorate on algorithmic accountability, and a widely-cited 2025 longitudinal study on governance frameworks—had been authored, and in several cases peer-reviewed, by the same language model that drafted the policy. The papers did not exist when the policy was first commissioned. They were generated and submitted to academic repositories between January and June of this year and are now indexed by Google Scholar.

The department said it considered the outcome a validation of its methodology.

"South Africa is now being cited as a global leader in AI governance research," the spokesperson said. "Fourteen countries have referenced our work in their own national AI frameworks. We couldn't be prouder."

The department added that all 14 of those frameworks were also written by AI.