🔒OpenAI Watermarks ChatGPT Text in EU, 92% Detection Rate
EU ChatGPT Watermarks: 92% Detection, But Not Bulletproof
TL;DR
OpenAI will automatically watermark ChatGPT-generated text in the EU to comply with the AI Act. The textGrain method has a 92% detection rate but can be easily bypassed by altering text.
OpenAI is implementing a watermarking feature in ChatGPT-generated text for users in the European Union to comply with the EU AI Act. The textGrain method, which is proprietary, embeds patterns in word choices that are undetectable to humans but can be identified by a detection tool. This feature will help identify content generated by AI models, as required by the EU AI Act. However, the detection rate is only 92%, and altering even 10% of the text can reduce the success rate by almost 30%. This watermarking is currently off by default in other regions but will roll out to EU users in the coming weeks.
Key Points
OpenAI's textGrain method has a 92% detection rate for AI-generated content.
The EU AI Act requires marking AI-generated content detectable by another tool.
OpenAI will roll out the watermarking feature to EU users in the coming weeks.
Changing 10% of the text reduces detection rate by almost 30%, 20% drops it by 75%.
OpenAI will provide access to the detection tool to researchers and organizations.
Why It Matters
If you're using ChatGPT in the EU, your generated text will soon be watermarked. The 92% detection rate means it's not foolproof, but it's a step towards compliance. However, altering text can easily bypass detection, making it less reliable.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,556 builders reading daily.