OpenAI Adds Text Watermarks in EU for AI Act Compliance
OpenAI is implementing text watermarking for ChatGPT and Codex in the EU.
"EU regulation is making AI transparent. OpenAI's watermarks are a big step, but editing can still make them tricky to detect."
OpenAI is rolling out invisible, machine-readable text watermarks for ChatGPT and Codex in the European Union. This move is to comply with the EU AI Act, which mandates that AI-generated content be identifiable by other systems. The watermarks will be available to eligible users on all plans within the EU, with developers globally able to activate them for select models via OpenAI’s API.
The watermarking method, called textGrain, subtly shapes the model's word choices to create a detectable pattern. OpenAI states that the watermark does not identify the user and does not significantly impact model performance. While editing can make detection harder, and short passages, math answers, and translated text are more challenging to watermark, OpenAI's tests show a 10% word replacement can drop detection from 92% to 66%. Initial detector access will be limited to approved researchers and expert organizations.
This initiative follows similar moves by other AI companies, with OpenAI previously having developed a text watermark but delaying its release due to concerns about users switching to non-watermarking rivals. OpenAI, along with Anthropic, Google, Meta, and Microsoft, has committed to adhering to the EU’s code of practice for AI-generated content.
This development highlights the increasing regulatory landscape for AI, particularly in the EU. Businesses using AI-generated content will need to be aware of provenance and compliance requirements, potentially impacting content creation workflows and verification processes.
Relevant tools
Find the right AI tool for your business
Chat with Insta and get matched to the right tool in seconds.
Try Insta Tool Finder →