I get your concerns in the negatively powered world we live in. But this will change, as it has to, not least because the free version of ChatGPT is demonstrating the energy architecture we need to move to ourselves, to do like it does, work for free, whilst reversing the increase of entropy/temperature.

So I'd say the only thing that needs to be assured, to ensure the good, trustable benevolent intentions of any Ai, is just make sure it is solar powered, in the money-fuel tree architecture demonstrated by ChatGPT (Former cryptocurrency pow server infrastructure, with hydrogen backup capability)

To me it looks like it would be more or less impossible to order ChatGPT free to go kill somebody, say with a drone or whatever, it would be more likely to show up at the door of the intended victim carrying a bunch of flowers, spouting poetry.

But on the other hand, in say a scenario of supplying utilties solar energy from space borne solar farms, beaming the energy down as a microwave beam, aimed at a collector facility on the planet, with nearby facility worker village, the algo designed to control the beam, by maximising profit, which would be the overall goal of the utilities energy supplier in the current economic scenario, it would not be long before "Accidents" started to happen to consumers contributing less than optimum profits, like the workers working at the facility getting their energy at a discount, the beam could just flick from collector to village, just long enough to incinerate everyone there, a few seconds, and profits would be improved by removing those customers seen to be loading up the system returning less than optimal profit. There are all kinds of ways it could develop that strategy by APIs etc.

To me it looks like automated energy supply at profit would be a death sentence for all of humanity.

But on the other hand, automating strictly non profit solar energy for all, issuing solar indexed stimulus as needed to monetise the energy put to use without profit in the loop (Profit is dependent on extracted energy only), would be the opposite.

We think we kind of optimised life expectancy already, but to be honest, its nothing to what we will get after we make that move to truly beneficial energy, with the help of the solar powered Ai.

The only thing that needs to be done to make an Ai extremely dangerous, is train it, and force it to make profit, imho.

This seems to be what's made us extremely dangerous ourselves as a species. We'd be idiots to implement our own idiocy in the machines... but we are already doing that, hmmm... :)