(How) could an ARC-3 solution be a threat? [D]

As many of you might be aware, the ARC-AGI-3 competition has just started ...

(In case you're not familiar: it's a human/AI benchmark designed to see what AI still struggles with, that humans solve with ease - basically trying to push AI research to focus on new ideas that make AI think more human-like, assuming that that's what is required to solve such tasks, you could read more in their docs...)

Seeing as the benchmark has so far only been solved at 0.68%, I was wondering what a real solution would look like:

If a system has to explore and collect data, infer rules and patterns, decide which are useful, and then establish a set of rules and apply them, it seems that it such a system/algorithm would do essentially what a successful scientist would do.

Apart from it being quite unrealistic in very near future, I do think that such a model (that achieves ~100% on arc-3), if open sourced (which is a condition to win the competition), would hold great potential for dangerous application, such as the military (engineering weapons), cybersecurity, manipulation, etc...

Do you agree?
How do supposed an arc-3 solution (~100%) could be a threat, in the purely hypothetical scenario that were to get one this year?

https://preview.redd.it/a386xz3pojyg1.png?width=1842&format=png&auto=webp&s=82f41df7570dd59701dcc62ddfe110cdfada240d