AI-detection tools promise objectivity but introduce new risks of error and bias.' Critically analyse in the context of newsroom AI governance.

UNESCO's Recommendation on the Ethics of AI (2021), adopted by 193 states, insists that accountability for published content stay with humans [4]. Detection tools like GPTZero and Pangram now mediate that duty in newsrooms — their objectivity is real but only partial.

The promise: scale and consistency

• Editors cannot hand-verify a rising volume of AI-written and AI-polished submissions; a probability score offers a uniform first filter.
• Pangram's transformer-based classifier, trained on large human and AI corpora, achieves substantially lower error rates than earlier tools [1].
• Screening supports UNESCO's newsroom asks of transparency and labelling of generative-AI content [3].

Risk of error

• Nature's editorial staff found human-reported journalism flagged as "100% AI", and AI-assisted translations wrongly flagged [1].
• To suppress false positives, vendors accept more false negatives — clearing genuine AI text [1].
• Crucially, a score cannot show how AI was used: a grammar fix and a fully generated piece look alike [1].

Risk of bias

• Detectors infer machine authorship from low perplexity — plain, predictable wording. That penalises a smaller vocabulary, not dishonesty.
• A Patterns study of 91 TOEFL essays by non-native English writers recorded a 61.3% average false-positive rate, with 97% flagged by at least one detector, while US school essays were judged near-perfectly [2].
• Enriching vocabulary cut misclassification to 11.6% [2] — the cheat escapes, the honest second-language contributor is caught. For Indian regional-language writers, this is an equity question.

Governance gap

• India's amended IT Rules cover synthetically generated information, but bind intermediaries and prescribe a 10% visual label — built for deepfake video, not a contributor's emailed prose [5].
• UNESCO's principles and the India AI Governance Guidelines (PIB, 2025) remain soft law [6][4].

Detection should therefore narrow who receives closer human scrutiny, never decide. Declaration of AI use at submission, a flagged writer's right of reply, and a Press Council code on the South-East European model [3] would align newsroom practice with UNESCO's core principle — human editorial accountability.