Poll: Does AI watermarking change how you would write an AI-use policy?

Anthropic started watermarking Claude's text output on 2 August. Not a visible mark, nothing you would ever see on the page. It is a pattern sitting under the words that software can read and a person cannot, and it travels with the text wherever you paste it. Worldwide, not just the EU.

Every education person I know read that as "finally, a detector."

It is not one. There is no public detection tool yet. A detected mark only means the text passed through Claude at some point, so grammar checking looks the same as generating the whole thing. And paraphrasing strips it, so no mark proves nothing either.

What I keep landing on is that this changes the evidence conversation, not the detection one. If you cannot prove who wrote it, you build the process so that the proof is in the drafts, the version history, and the person explaining their own work out loud.

Curious where everyone else sits on this in their own sector.

Yes, once detection tools ship I would build them into the policy

No, evidence of learning and process is the only thing I would rely onc

Holding off until I see how accurate detection actually is

Honestly, we do not have an AI-use policy yet

2 votes

2 comments

Poll: Does AI watermarking change how you would write an AI-use policy?