OpenAI is beginning to add invisible, machine-readable watermarks to text generated by ChatGPT and Codex, starting with users in the European Union.
The new system, called textGrain, is initially being rolled out to eligible ChatGPT and Codex users across all plans in the EU. OpenAI says the technology is designed to help identify whether text was generated by its AI models, although it will not be enabled as a global default for now.
OpenAI says textGrain performed as well as or better than other text-watermarking approaches, including Google DeepMind’s SynthID. The company is introducing the technology as part of its efforts to improve transparency around AI-generated content and meet requirements under the EU’s AI Act.
However, OpenAI is stressing that the watermark should not be treated as definitive proof that a piece of text was created by AI. The company says textGrain cannot reliably detect every watermark and does not establish who owns a piece of text, how much a human contributed, whether the information is accurate, or whether a person wrote the content themselves.
The company is also making watermarked text available as an opt-in feature for API customers worldwide. Starting now, customers using select models can choose to watermark generated text based on their own transparency requirements. OpenAI also says it is working with cloud partners to bring the technology to model outputs accessed through their services.
OpenAI is also opening access to its watermark detection system for approved researchers and expert organizations. Applications are now available, with access initially being provided on a case-by-case basis to support research and improvements to text provenance technology.
The detection tool will only indicate whether an OpenAI watermark is detected. OpenAI says it will not reveal the user’s identity, prompts, or conversations. The company is also keeping the detector away from the general public for now because watermarks can sometimes be missed or incorrectly detected.
For users in the EU, the text watermarking rollout will happen over the coming weeks. OpenAI says it is taking a regional approach so it can learn from real-world use and feedback before deciding whether the technology should become more widely available.
If this article helped you, please consider supporting our work. Every small contribution keeps Abijita.com independent and running.
