OPENAI IS ADDING INVISIBLE WATERMARKS TO CHATGPT AND CODEX FOR USERS IN THE EU

Summary from the TLDR Design newsletter

OpenAI will begin adding invisible, machine-readable watermarks called textGrain to eligible ChatGPT and Codex output in the EU to comply with transparency requirements in the EU AI Act. The technology embeds a statistical pattern into generated language rather than adding hidden characters. API customers worldwide can optionally enable it for supported models. OpenAI acknowledges that watermarking remains imperfect because editing and short outputs can weaken the detectable signal.