Stories
OpenAI Introduces Hidden Text Watermarks for ChatGPT and Codex in the EU
F
Firstpost•06-10-2026, 07:46
OpenAI Introduces Hidden Text Watermarks for ChatGPT and Codex in the EU
- •OpenAI is embedding invisible signatures in text generated by ChatGPT and Codex for EU users to comply with the EU AI Act.
- •The new technique, named textGrain, alters word choice probabilities to create a statistical pattern detectable by authorized researchers.
- •Global developers using OpenAI's API can manually enable text watermarking for select models worldwide.
- •Modest edits like replacing 10 percent of words with synonyms can reduce the detector's confidence significantly.
- •OpenAI restricts general public access to the detector initially, providing it only to approved researchers and specialist organizations.
Key Questions Answered
OpenAI has named its text watermarking technique textGrain. The method alters the probabilities used when the model selects its next words to create a statistical pattern that a detector can search for.
OpenAI is introducing the hidden watermark to comply with transparency provisions under the EU AI Act, which took effect on August 2 and require AI-generated material to be identifiable by other systems.
OpenAI is initially restricting access to its detector to approved researchers and specialist organisations so they can test how dependable the technology is and where it can be used responsibly.
