I think with OP’s analogy, a nuclear weapon is no longer a tool, because the actual tool (nuclear technology) was crafted into something (weapon) that does a specific thing (destroy). But I guess OP should’ve used LLM and machine learning instead of AI.

That's a great example actually, you just have to differentiate between a nuclear weapon and a nuclear power plant. You can split atoms to build a bomb or heat water, the concept of splitting an atom itself isn't evil.

Nuclear fission has been used in nuclear reactors for good, and was exceedingly efficient at that time, and has been a significant factor in the recent human history. Just as it's been with the "evil" use in nuclear weapons.

I'm not aware of any breakthrough that has not been used for both good and evil.

Good justification to why GOS is right to use AI to detect vulnerabilities. The only true way to protect yourself from an enemy that has nuke bombs is to have some yourself.

Ukraine gave up its stockpile in exchange of promises from both Russia and NATO that no harm would come to them, and would there be harm from one side, the other side would come and help them. They were betrayed by both sides.

AI is very efficient at finding vulnerabilities. Among a see of false positive, they find real ones. It's much more efficient and faster than human auditing code. So either it's GOS using it, finding them and patching them, or it's… someone else.