Overview
This is to comply with the region's new AI transparency rules.
Overview
OpenAI announced plans to comply with part of the European Union's new AI transparency rules by adding an invisible watermark to text and code generated in the EU, following a similar move by Anthropic.
Technology and Detection
OpenAI is using a proprietary system called textGrain that embeds an invisible statistical signal into the model's word choices, allowing a specialized detector to verify if a watermark is present.
Technology and Detection
The company states that the system matched or exceeded alternative approaches like SynthID, though it cautions that reliable detection is not guaranteed under all everyday conditions.
Limitations and Open Source
Shorter text and mathematical content remain harder to detect, yielding lower success rates, while edited text also reduces detection accuracy. OpenAI plans to open-source the technology for external refinement.
The watermark will be enabled by default for eligible ChatGPT and Codex text outputs in the European Union, while customers can opt in for select models.
Access to the detection software will initially be restricted to approved researchers and expert organizations without exposing user identities, prompts, or conversations.
Article 50 of the EU AI Act requires generative AI providers to make text outputs machine-readable and identifiable, giving pre-existing market entities a strict deadline to comply alongside companies like Microsoft, Google, and Meta.
