🔒OpenAI Adds Invisible Watermark to ChatGPT and Codex in EU
EU AI Act forces OpenAI to watermark ChatGPT and Codex
TL;DR
OpenAI is adding an invisible watermark to ChatGPT and Codex in the EU to comply with the AI Act. Developers can opt-in globally, but it's off by default. The watermark subtly shapes word choices, detectable by a tool.
OpenAI is rolling out an invisible watermark to ChatGPT and Codex in the EU to comply with the EU AI Act. Developers can turn it on globally, but it's off by default. The watermark subtly shapes word choices, detectable by a tool but not visible to users. It doesn't identify the user and doesn't affect model performance. The EU AI Act requires AI companies to mark AI-generated content in a way other systems can identify. OpenAI's textGrain method subtly shapes word choices, making detection possible. The watermark travels with the text when copied and pasted, but it can be removed by editing.
Key Points
OpenAI will roll out the watermark to ChatGPT and Codex in the EU over the coming weeks.
Developers can turn on the watermark for select models starting today, globally, but it's off by default.
The watermark subtly shapes word choices, detectable by a tool but not visible to users, leaving a pattern readers can't see.
OpenAI's textGrain method doesn't affect model performance and travels with the text when copied and pasted.
Removing the watermark by editing drops detection from about 92% to 66%, making it less reliable for short passages.
Why It Matters
If you're developing AI applications in the EU, the invisible watermark on ChatGPT and Codex affects your compliance strategy. The watermark's subtle nature means it's detectable but not visible, impacting transparency and trust. Developers should consider the implications for their applications' transparency and user trust.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,556 builders reading daily.