What was announced
OpenAI's October announcement distinguishes a planned EU rollout for eligible ChatGPT and Codex text from optional watermarking for selected API models. Its account also describes false positives, missed detections and reduced detection after editing. Availability and scope should be checked against the dated announcement, rather than assumed to be identical across all products.
Detection is not a verdict on a person
For a teacher, editor or manager, the tempting shortcut is to turn a detector result into a judgment about authorship. That is a larger conclusion than the signal supports. A watermark does not establish truth, ownership or the amount of human contribution. An absent signal is not proof that a human wrote the whole passage. The distinction matters when someone could face criticism or sanctions.
Ask a better editorial question
Instead of relying on a single score, ask what claims the text makes, whether its sources support them and how the author checked the result. Where a publication or institution has disclosure rules, those rules need to be explained separately. Technical detection and an organization's editorial expectations answer different questions.
A correction to the practical advice
The earlier version suggested considering other providers that might not face the same regulations. Provider switching is not a demonstrated way to resolve a legal obligation, and that advice has been removed. Specific legal duties require separate analysis of the applicable rules and use case. For readers, the immediate task is to understand what a detection result actually says before using it to judge a text or its author.