OpenAI has described its approach to the EU’s requirements on the provenance of AI-generated content. The company plans to use an invisible technical signal for eligible text outputs to help tell whether material is linked to its tools. At the same time, OpenAI itself stresses that such a signal doesn’t replace fact-checking and isn’t absolute proof of authorship.
Why text is harder to label than images
Text is easy to rewrite, shorten, translate or mix with passages written by a person. After such changes, the technical marker can disappear or stop being reliably detected. That’s why labeling is useful as an extra hint, but not as the only criterion in moderation, grading or authorship disputes.
- the signal is meant to help identify where a text came from;
- editing and translation can make detection worse;
- the absence of a signal doesn’t prove AI wasn’t used;
- the presence of a signal doesn’t replace a human assessment of the content.
For readers, the practical takeaway is simple: treat labeling as context. Check the author, the date, the original source and the facts — especially in news, financial advice and medical content.
FAQ
Can any AI-generated text be detected reliably?
No. Technical methods have limitations and can lose reliability after editing, shortening or translation.
Does the absence of a label mean a person wrote the text?
No. The absence of a technical signal isn’t proof of human authorship.
