Submission to the European Commission on the Draft Guidelines on Article 50 AI Act: Distinguishing AI-Generated Public-Interest Text from AI-Assisted Human Authorship

This submission concerns paragraphs 122–128 of the Draft Guidelines, addressing Article 50(4) subparagraph 2 AI Act: AI-generated or manipulated text published for the purpose of informing the public on matters of public interest, and the exception for text subject to human review or editorial control and editorial responsibility. Lex et Ratio Ltd welcomes the Commission’s objective of reducing impersonation, deception, misinformation, manipulation at scale, fraud, and harm to democratic processes, societal trust, and the integrity of the information ecosystem. The submission is limited to one proposed clarification. The final Guidelines should distinguish more clearly between AI-generated public-interest text and human-authored public-interest text produced with AI assistance. Disclosure under Article 50(4) subparagraph 2 should turn on a materiality threshold consistent with the approach already adopted in paragraphs 84–86 on standard editing and non-substantial alteration, and paragraph 109 on minor manipulations that do not transform content into a deep fake. The central point is that AI assistance is not the same as AI authorship. Where a human author determines the thesis, selects the sources, verifies the claims, controls the reasoning, approves the final text, and accepts responsibility for publication, the work is better characterised as AI-assisted human authorship, not AI-generated publication. The principal drafting ambiguity arises at paragraph 127. Its exclusion of ‘superficial, solely formal or procedural checks’ is appropriate, but could be read too broadly unless the final Guidelines also clarify what substantive human review does not require. Substantive human review should not require that every sentence be manually drafted without AI assistance, nor that every AI-assisted formulation be rewritten from scratch. The requested clarification is that auxiliary AI use does not, of itself, trigger the disclosure obligation under Article 50(4) subparagraph 2 where the final publication has undergone substantive human review or editorial control and is subject to editorial responsibility. Disclosure should remain required where AI has materially generated or transformed the published text, or where non-disclosure would mislead the intended audience about authorship, provenance, expertise, accountability, or evidential basis.