I only saw Across The Spider-Verse in a Theater, but oh my, one of the most must-see-in-an-actual-cinema movies of the last 20 years. Absolutely immersive.

Maybe it's actually a good way?
Sure it's a bad way to sell AI, and misleading of how much better it can already be.
But I think it serves as a reminder of the subtle stupidity behind even the most advanced models. :-)

The specific text "Signatures using elliptical curve cryptography (ECC), such as ECDSA, aren't supported in Windows and newer Windows security features." seems pretty clear to me, "It's not supported".

...

* Digicert claims (docs.digicert.com/en/certcentr...) that we could use ECC P-256-bit key size.
* Yubico timings (support.yubico.com/hc/en-us/art...) indicate that ECDSA-P256-SHA256 could be more than facto 10 faster than RSA-4096. (Factor 6 for RSA-3072)

The reason I wanted to try ECC:
* Signing all binaries of our CI release builds is dead-slow with RSA 4096 (850 msec per File on our YubiHSM2), and ...

In my case we would have liked to try an ECC only private key for our OV certificate, so the Root certificates would still be RSA.
Seeing ECC unsupported on the Root however is not exactly encouraging.

Interesting. Even though the docs you have linked to are specifically for the "Trusted Root", which, as far as I can grasp, I don't have any influence over anyway with Authenticode, since that's provided by (e.g.) Digicert.