OpenAI is rolling out an invisible, machine-readable watermark known as textGrain for text generated by ChatGPT and Codex. This feature launches exclusively for users within the European Union to comply with the EU AI Act. The company states that this method matches or exceeds the performance of Google DeepMind’s SynthID, a system also adopted by Anthropic. OpenAI published benchmark scores showing that watermarked text retains similar quality to unwatermarked output, though it explicitly notes that the technology does not guarantee detection by all scanners. While the move satisfies regulatory requirements for provenance, it has not been universally welcomed by the wider community.
The implementation serves as a direct response to legislation mandating transparency for AI-generated content, yet the practical utility remains limited to specific jurisdictions. The primary value lies in providing a technical standard for regulators rather than offering immediate consumer verification tools.
* Launch is restricted to the European Union only
* TextGrain aims to match SynthID detection rates
* OpenAI claims no loss in generation quality
