OpenAI adds text watermarking to ChatGPT for users in the European Union

OpenAI is introducing machine-readable digital watermarking for text produced by ChatGPT and Codex. This rollout initially targets users located within the European Union to satisfy specific mandates established by the EU AI Act. While the company is making this feature available to API customers for global content, the automatic application of these marks is currently restricted to the European market. OpenAI utilizes its proprietary textGrain technology, which it claims performs at levels comparable to Google DeepMind's SynthID.

This development is a strategic response to increasing regulatory pressure regarding content provenance and the prevention of misinformation. By making AI-generated text identifiable to machines, OpenAI aims to comply with emerging legal frameworks. For the corporate sector, these changes could necessitate adjustments to internal workflows, particularly concerning how organizations generate, transmit, or document official records such as tax filings and invoices. Although the technology aims to improve transparency, the company acknowledges that these watermarks do not provide an absolute guarantee of detection.