OpenAI has begun deploying a specialized text watermarking framework across its artificial intelligence platform, initiating the rollout exclusively for accounts located within the European Union. The invisible, machine-readable system embeds markers directly into text output generated by both ChatGPT and the code-focused software Codex.
The underlying technology, designated by OpenAI as textGrain, introduces subtle patterns into model outputs that are completely undetected during standard human reading but remain recognizable to automated detection software. Internal evaluations conducted by OpenAI showed that textGrain matched or exceeded the performance of alternative industry approaches, including Google DeepMind's established SynthID system for text verification.
SynthID has rapidly become a central baseline for artificial intelligence watermarking efforts across competing laboratories. In August, rival AI developer Anthropic announced its own text watermarking initiative, explicitly utilizing Google DeepMind's SynthID framework to power its content detection pipeline.
What it means
The decision to restrict the initial availability of textGrain to the European Union reflects the region's prominent position in driving artificial intelligence transparency and content attribution standards. Technology providers are increasingly tailoring product features to meet European expectations surrounding the identification of synthetic media.
By deploying textGrain across both conversational software like ChatGPT and technical development utilities like Codex, OpenAI is establishing automated watermarking across multiple distinct categories of text generation. How textGrain fares in live European deployments compared to established alternatives like SynthID will likely influence how machine-readable watermarks are adopted worldwide.
