Then why release this outdated version?

The mitigations don't seem to address targeted attacks that I'd think you can assume will happen.

How do you address the civil and criminal liability of that?

Are there even any criminal laws against this? I mean, it strikes me that there should be, but my non lawyerly self has never heard of any.

Yes, absolutely. Using one‘s likeness without explicit consent first is illegal in (most of?) Europe and is a tricky subject in the US. For example look at Crispin Glover‘s lawsuit against Back to the Future II.

https://en.wikipedia.org/wiki/Personality_rights

To be clear, I'm looking for references to criminal law, not civil or case law. This all seems a combination of the latter, and even here it's not obvious what applies in cases where a third party produces the infringing content.

I am not a lawyer.

I had to briefly look up the difference between criminal law and case law to see what you mean. I have no idea if there are any criminal cases in the US about this.

In Civil Law countries as opposed to Common Law countries the plaintiff could very well be the state for this type of legislation. For example, look at tech companies being fined for GDPR violations, same basic idea.

Your mitigations are so bad, you must know that they are almost useless. No website has ever kept up with spam by individually checking every complaint, and you would only stop people after they've already recorded large amounts of output from a stolen voice.

On the other hand, as you also know, you could easily put a much stronger safeguard on by making people say a short prepared statement and only clone voices who have said that statement. You do not appear to have done that. Why not, except it would make it harder to voice steal?