OpenAI will begin adding invisible watermarks to text generated by ChatGPT and Codex in the European Union over the coming weeks, the company announced Monday.
The move comes as the EU's AI Act requires providers of generative artificial intelligence systems to make AI-generated text identifiable in a machine-readable format.
OpenAI said its technology, called textGrain, embeds an invisible statistical signal into the words selected by its models. A detector can then look for the signal to assess whether a passage contains an OpenAI watermark.
The company cautioned that text watermarking and detection remain early-stage technologies with significant limitations.
Shorter texts are harder to detect, while editing, paraphrasing or translation can weaken the watermark.
OpenAI also stressed that detecting a watermark does not prove a text was entirely AI-generated, while its absence does not prove it was written by a human.
The company is opening applications for access to its watermark detector, initially limiting to approved researchers and expert organizations.
OpenAI said the phased rollout reflects both the requirements of the EU AI Act and the limitations of current watermarking technology.
