📌 ChatGPT Will Start Adding Hidden Watermarks to Text in the EU

ChatGPT and Codex will start adding watermarks to responses in the EU in the coming weeks. This won't become the global default. Watermarks can also be enabled for select models through the API, but they'll be off by default.

OpenAI notes that watermarking may work less well on short texts and in subject areas where there's less flexibility in wording, and that manual edits can weaken the signal.

For now, the detector will be available only to approved researchers.

Should AI-generated texts be labeled?

❤️ — Yes, that's the right thing to do!
👎 — No, it'll just get in the way