By 2026, the initial panic over generative AI in higher education has led to a standoff, with major South African universities switching off AI detectors. Dr Mario Landman of the Academic Centre of Excellence for The IIE and Advtech cites a lack of accuracy as the primary reason, noting that AI detectors measure statistical signatures like "perplexity" and "burstiness," which have become blurred as generative models evolve. Independent evaluations show that while some tools claim 99% accuracy, effectiveness drops to between 60% and 80% when students manually edit or use "humanise" prompts. Newer models like Claude 3 frequently evade mainstream checkers. For administrators, using probabilistic tools for disciplinary decisions risks due process. The most damaging aspect is documented bias against non-native English speakers; research shows detectors disproportionately flag ESL students because their formal writing is mistaken for machine-generated patterns. A landmark study found a 61.3% false positive rate for TOEFL essays by Chinese students, compared to 5.1% for native speakers. In multilingual South Africa, this creates a systemic equity crisis. Dr Landman also notes a "devil’s bargain" where AI automates tasks, producing "shallow knowledge workers" deficient in critical analysis. The emerging way forward is a shift from "policing" to "stewardship."