OpenAI is adding text watermarking in ChatGPT and Codex
An invisible, machine-readable watermark in text output is rolling out to ChatGPT and Codex, but only for users in the European Union at first. OpenAI says its textGrain watermarking "matched or exceeded" other approaches like Google DeepMind's SynthID for text, which is also the basis for the…
OpenAI is implementing text watermarking in ChatGPT and Codex, initially for EU users, to address regulatory requirements and enhance transparency.
What happened
OpenAI is introducing text watermarking in ChatGPT and Codex, starting with EU users. The watermarking technology, textGrain, is designed to be machine-readable and invisible. It is part of OpenAI's efforts to comply with the EU's AI Act and provide transparency. The feature is not yet a global default and is available for API customers with an opt-in option. Researchers can apply for access to detector tools to evaluate text provenance.
Why it matters
Text watermarking is significant for AI developers and businesses as it addresses regulatory compliance, particularly the EU's AI Act. It enhances transparency and helps in tracking the origin of AI-generated content, which is crucial for maintaining trust and accountability in AI systems.
- Text watermarking is a step towards greater AI transparency and regulatory compliance.
- OpenAI's approach is regionally phased, starting with the EU.
- Researchers and businesses can access tools to evaluate and implement watermarking.
- Watermarking does not verify text accuracy or authorship, highlighting its limitations.
- The technology is not a global default, allowing for flexibility in adoption.
The bigger picture
The implementation of text watermarking in ChatGPT and Codex impacts AI transparency and regulatory compliance. It allows for better tracking of AI-generated content, which is essential for meeting legal requirements and maintaining user trust. However, the technology's limitations, such as not guaranteeing reliable detection, mean that additional measures may be needed for comprehensive verification.
OpenAI is introducing text watermarking in ChatGPT and Codex, primarily for EU users, to enhance transparency and meet regulatory requirements. The feature is being rolled out in phases, with options for API customers and researchers to access and evaluate the technology.
Technical details
- textGrain watermarking technology
- machine-readable and invisible watermarks
- AI benchmarks for performance comparison
- opt-in watermarking for API customers
- detector tools for researchers and expert organizations
FAQ
What is text watermarking in ChatGPT and Codex?
Text watermarking is an invisible, machine-readable mark added to AI-generated text to help identify its origin and enhance transparency.
Why is OpenAI implementing text watermarking?
OpenAI is implementing text watermarking to comply with the EU's AI Act and improve transparency in AI-generated content.
Is text watermarking available globally?
No, text watermarking is initially available only for users in the European Union and is not a global default at launch.
Can researchers access the watermark detection tools?
Yes, approved researchers and expert organizations can apply to access the detector tools for evaluation purposes.
What are the limitations of text watermarking?
Text watermarking does not guarantee reliable detection, verify text accuracy, determine ownership, or prove human authorship.
This summary and analysis were generated with AI from the original article and may contain errors. Read the original for the full story.
Read the full story
Originally published by The Verge - AI on Oct 5, 2026.
Comments
Sign in to join the discussion.