OpenAI is embedding a machine-readable watermark technology called textGrain into ChatGPT and Codex. The first batch of users to "enjoy" this feature are those in the European Union.

OpenAI says this watermarking scheme is not inferior to Google DeepMind's SynthID, and is even better. Competitor Anthropic has already announced in August this year that it would adopt SynthID-based watermarking technology. Everyone is rushing to do the same thing: comply with the EU's AI Act.

The watermark is embedded in the generated text and can be read by machines to determine whether the content was generated by AI. The key point is that OpenAI says the performance of text generation with the watermark is basically the same as without it.

Currently, this technology primarily targets EU users; other regions have not been mentioned for now.

Official announcement: https://openai.com/index/eu-text-provenance/