A human caught the model.
In April 2026 review found an AI session had fabricated or paraphrased part of a BSBTEC301 performance-evidence requirement. The error was checked against the external authority and the source-authority rule was strengthened. Oversight working, not system perfection.