00:00Using AI to write? Your text could soon carry a hidden label.
00:04Starting this month, Anthropic is adding invisible watermarks to texts created by its newest AI models.
00:10AI-written essays, blog posts, and even flirty DMs could carry a hidden AI label that only machines can read.
00:18Anthropic says it will apply the watermarks globally and as developing tools that will allow people to check whether a
00:23text contains one of Claude's hidden watermarks.
00:26This comes after the EU AI Act, which requires all AI-generated content to be marked.
00:32So, how does it work?
00:33The hidden system subtly steers Claude's word choices into a detectable pattern.
00:38You won't notice it, but specialized detectors can.
00:41Supporters say this could help enforce no AI policies in settings like academic conferences.
00:46But watermarks aren't perfect.
00:48While simple copy and pasting won't remove them, heavy rewriting or translation can.
00:53And a watermark alone doesn't prove how AI was used.
00:57What do you think?
00:58Will text watermarks restore trust in the written word?
01:01Or discourage people from using helpful AI tools?
Comments