● Covered by 4 sources · 4 reportsMedium impact4 neutral
OpenAI Introduces textGrain Watermarking for AI-Generated Text in EU and API
Key points
- OpenAI introduced textGrain, an invisible digital watermarking system.
- It embeds a statistical signal into AI model word choices for detection.
- Automatically applied to ChatGPT and Codex in the EU due to EU AI Act.
- Available as an opt-in feature for API customers worldwide.
- Detection reliability varies with text length and content type.
New Watermarking System for AI Content
OpenAI has announced the implementation of a new proprietary system called textGrain, designed to add an invisible digital watermark to text and code generated by its artificial intelligence models. This system works by embedding a statistical signal into the model's word choices, subtly influencing the selection of words to create a detectable pattern. A detector can then identify these signals to ascertain if a watermark is present.
Compliance with EU AI Act
The introduction of textGrain is a direct response to the European Union's new AI transparency rules, specifically the EU AI Act, which took effect on August 2. This legislation mandates that providers of generative AI systems mark AI-generated content in a way that allows for identification. OpenAI will automatically apply this watermarking to eligible text produced by ChatGPT and Codex for users within the European Union in the coming weeks.
Global API Availability
Beyond the EU, OpenAI is making textGrain available to API customers globally. Developers using OpenAI's API can opt-in to use the watermarking feature for select models starting immediately. However, the watermarking will remain off by default for API users, allowing them to choose whether to implement it.
Detection Capabilities and Limitations
OpenAI states that textGrain's performance 'matched or exceeded' other approaches, such as SynthID for text. The watermark is not a visible symbol but rather a pattern within the words themselves, meaning it travels with the text if copied and pasted. OpenAI also noted that the watermark does not identify the user and did not cause a meaningful change in model performance.
However, the company cautions that 'strong performance under ideal conditions does not guarantee reliable detection in everyday use.' Shorter texts, as well as content like mathematics, are harder to detect, with an approximate 80 percent success rate. Editing the text can also reduce the success rate of detection. OpenAI plans to open-source the technology for further development.
Industry Context
OpenAI's move follows a similar initiative by Anthropic earlier this year, which also announced plans to watermark text generated by its Claude model to comply with the EU AI Act. Anthropic's system also relies on subtly adjusting word choices to create a detectable pattern. OpenAI's technical report for textGrain was co-written with researchers from a university, indicating a collaborative approach to its development.
✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →
The daily brief
One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.
One email a day. Unsubscribe in one click, any time.
Today's brief
Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.
▶ Play today's brief
New every morning, and the back catalogue is archived by date.
Primary sources
GitHub facebookresearch/textsealHow outlets covered it
OpenAI introduced text watermarking for its generative AI models, including ChatGPT, Codex, and API, to comply with the EU AI Act. This system embeds an invisible statistical signal into word choices, similar to Anthropic's approach, allowing for detection of AI-generated text.
OpenAI will implement an invisible watermark on text generated by ChatGPT and Codex for users in the European Union. This change is in response to the EU AI Act's transparency rules, which mandate identification of AI-generated content.
OpenAI has launched textGrain, a new watermarking system for text generated via its API, which embeds a statistical signal by subtly influencing word choices. This feature is opt-in for API customers globally but will be automatically applied to eligible text from ChatGPT and Codex in the EU in response to the EU AI Act.
OpenAI will implement an invisible digital watermark, called textGrain, on text and code generated within the European Union to comply with the EU's new AI transparency regulations. This system adds a statistical signal to model word choices, allowing for detection, though its reliability varies with text length and content type.