AI Watermarking: We Comply—and We Tell You More

Regulation (EU) 2024/1689, Article 50, creates transparency duties for certain AI systems and uses. Providers of systems that generate synthetic audio, image, video, or text content must mark outputs in a machine-readable format and make them detectable as artificially generated or manipulated where technically feasible, subject to stated limitations such as standard editing and inputs that are not substantially altered.

Deployers have different duties. They must disclose covered deepfakes, and must disclose AI-generated or manipulated text published to inform the public on matters of public interest unless the text has undergone human review or editorial control and a natural or legal person holds editorial responsibility. Other Article 50 duties cover direct AI interaction, emotion recognition, and biometric categorisation.

This page is public information, not legal advice. Applicability depends on the system, role, content, publication context, jurisdiction, exceptions, transition rules, and current governing text.

Veristio Press supports honest disclosure and opposes generalized labels that flatten very different creative facts into the same public signal. Grammar correction is not the same as full generation. Summarizing source material is not the same as fabricating a realistic event. Prompting alone does not establish the same authorship facts for every work.

Veristio Press objects to compulsory, overbroad, blunt, or provider-controlled marks when they stigmatize lawful work, erase human contribution, interfere with creative presentation, or make unverifiable technical assertions. Required or provider-applied marks may coexist with Veristio's fuller disclosure; our mark does not replace a legal duty.