Termsslug: llm-watermarking

LLM Watermarking

大模型水印

Technique embedding invisible statistical features in generated text to identify its source.

Curated Intelligence

1 items · Page 1/1
Ars Technica AIT2·Models
68 pts
OpenAI Enables Default Text Watermarking for ChatGPT in EU

To comply with the EU AI Act, OpenAI has announced that ChatGPT outputs will be watermarked by default in the European Union using a proprietary method called 'textGrain,' while remaining opt-in elsewhere. The watermark embeds invisible patterns in word choice distributions to enable machine detection, though technical documentation acknowledges it is easily circumventable by users with basic know-how. Detector access is currently restricted to approved researchers and organizations.