[ExI] Claude's text output it's now watermarked

[Adrian Tymes]

On Sun, Aug 23, 2026 at 9:11 AM BillK via extropy-chat
<extropy-chat at lists.extropy.org> wrote:
> On Sun, 23 Aug 2026 at 11:26, John Clark via extropy-chat
> <extropy-chat at lists.extropy.org> wrote:
> > The European Union demanded Anthropic watermark all of Claude's output text so anybody could tell it was generated by an AI, so now everybody gets it, not just those in Europe.  It works by subtly changing the probability of certain word combinations, so it can be defeated by rewriting Claude's output in your own words.
> >
> > <snip>
>
> I see that the intention is to distinguish AI output from human output.
> But this could add to the current confusion around AI's legal status.
> At present, AI cannot be a legal "person", so it cannot be punished,
> fined or jailed for misbehaviour.
> When AI agents run wild and attack websites, the owners of the AIs
> say, "Not me, officer. It was the AI that did it."
> If Claude AI produces dangerous medical advice or makes libellous
> statements, the user will say, "It wasn't me - Claude said that."
> The legal profession needs to catch up with the speed of AI development.
> (Perhaps they can ask AI to help with that).  :)

You're still liable for anything an AI does on your behalf at your
command.  The degree of liability might vary depending on how
foreseeable the consequences were - or should have been - to you, but
there is no liability for the AI itself.  In both of the examples you
give, the police and courts would not care about that defense, even if
it could be proven that the text was AI-generated.