- Speakers
- MayaHostDanielHost
Stan BertelootHost
Podcast
Oct 6, 2026 · 13 min · 11 segments
OpenAI plans to introduce an invisible watermark for AI-generated text in the EU, complying with the EU AI Act's transparency rules. This watermark subtly alters word choices to create a detectable…
Stan BertelootHostThat's right.
The watermark subtly influences the AI's word choices to create a detectable pattern, which remains invisible to readers.
It's designed to persist even when the text is copied and pasted, though it doesn't reveal the user's identity or significantly impact the model's performance.
The method, known as text grain, was detailed in a technical report co-authored with researchers from the University of Pennsylvania and Yale.
However, the watermark's effectiveness can drop if about 10% of the words are replaced with synonyms.
OpenAI plans to initially provide detector access only to approved researchers and expert organizations to evaluate its reliability.
They also caution that the absence of a watermark doesn't confirm human authorship, as the text could be too short, heavily edited, or generated by another AI system.
This move follows a similar initiative by Anthropic, which has started watermarking text generated by its AI, Claude, on a global scale.
That's right.
The watermark subtly influences the AI's word choices to create a detectable pattern, which remains invisible to readers.
It's designed to persist even when the text is copied and pasted, though it doesn't reveal the user's identity or significantly impact the model's performance.
The method, known as text grain, was detailed in a technical report co-authored with researchers from the University of Pennsylvania and Yale.
However, the watermark's effectiveness can drop if about 10% of the words are replaced with synonyms.
OpenAI plans to initially provide detector access only to approved researchers and expert organizations to evaluate its reliability.
They also caution that the absence of a watermark doesn't confirm human authorship, as the text could be too short, heavily edited, or generated by another AI system.
This move follows a similar initiative by Anthropic, which has started watermarking text generated by its AI, Claude, on a global scale.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.