What this is
The EU AI Act requires AI-generated text to be traceable to its source. "Text watermarking" — embedding invisible markers in AI output that algorithms can detect — is the most-discussed technical approach. This week, OpenAI revealed its implementation details: which texts receive watermarks, how the detection mechanism works, and one critical decision — initial detection access goes only to researchers, not to general users.
Industry view
Supporters see this as the first public compliance sample, giving the industry a referenceable engineering standard. But we see several reservations worth noting. First, restricting detection tools to researchers means journalists, teachers, and content platforms — those who most need to identify AI-generated content — cannot access them. Second, watermarks can theoretically be stripped or bypassed, putting regulatory effectiveness in question. Third, this standard is defined and enforced by OpenAI itself, with no neutral third-party oversight. One side effect worth watching: once watermarking becomes a compliance baseline, smaller model companies that haven't adopted traceability systems could get pushed to the margins.
Impact on regular people
For enterprise IT: Those handling EU business need to bring AI output compliance into their audit scope. "Detectability" will become a new procurement criterion when evaluating AI tools.
For professionals: The experience of using ChatGPT to write reports and emails won't change in the short term, but "content source traceability" will gradually be written into company internal policies.
For consumer markets: General users won't have access to detection tools for now, but social platforms, news aggregators, and educational content platforms are likely the first scenarios to integrate detection capabilities.