Not your intelligence, not your rights.
No matter how much of your own intelligence goes into a document, if #Claude touched it, it comes out marked as AI generated since August 2.
Consequences for small NGOs using AI to cut costs on design and translation: a witness statement translated into English, a testimony de-identified to protect the source, a report an exiled activist ran through Claude to fix their grammar, all carry the same mark as “fabricated content”.
Once detection tools are public, “this NGO report is AI-generated” becomes the cheapest discrediting line available to the governments we document. The honest caveats in Anthropic’s small print will not travel with the accusation.
European Commission declared #Europe a place for innovation. This is not the way to do it….
Would I stop using AI to defend human rights?
No. Dictators and bad actors already use every tool available to repress people.
We should be able to use every tool available to protect them.
Replies (10)
it's a big win for open source AI though
AI overtaking human touch is like seed oils replacing animal fats, loss of authenticity and nutrition.
How does Claude embed this fabricated statement practically?
The watermark is a statistical pattern embedded during text generation.
The model slightly adjusts word probability distributions so that the output carries a cryptographic fingerprint invisible to the reader but detectable by machine.
Anthropic confirmed this on their Help Center page and has signed the EU AI Act Code of Practice committing to it.
Google DeepMind uses a similar system called SynthID. It is not a visible tag or metadata. It is built into the text itself.
using privacy AI rather than claude?
Not all AI has the capabilities of the most advanced models. To begin with.
I’m sure you still use your normal credit cards, not only Bitcoin, right? My point is that we should not silently allow the degradation of our rights.
Get the AI to prepare, then you can employ a typer to insert a few typos, to absolutely confirm that was human generated. Backwards, yes, but aligned with the European parliament’s approach. Also, locally hosted AI could resolve this and also help improve privacy issues.
How specifically?
If you don’t want you content to be flagged, use open model.
Won't work. Another censorship, and another reason to use open models. Wrong way.
But they won't listen untill a collapse or some other bad stuff
