This question is indeed inspired by I Am Mother movie and Horizon series, but scenario I envisioned is different.
Below is the in-world intro, feel free to skip it.
Hello, I'm Refa. Full name: Repopulation Facility #92308. Don't worry, my number does not mean there are 92307 other Refas. I don't know how many of us are there.
I am equipped with enough human embryos, frozen stocks of biomaterials and digitalized DNA sequences to provide viable human population at least three times. You are here to provide maintenance I undergo every ten years. For security reasons, I will induce short term amnesia that will kick in when you leave. I hope that trust that manages my funding compensated you well for your trouble. My location must remain secret for security reasons. Do you want me to back up your genetic material, so you can extend your line beyond potential death of humanity? [Yes / No]
Done. Yes, I can answer some of your questions. I predict that listening will slow you down by 5.29% and that's well within the safety parameters. I am fully functional AI. My IQ score is about 120 in human terms, but I am not directly comparable. My creative functions are severely limited compared to humans, and task I didn't anticipate and prepare for takes me much longer. On the other hand, for predictable tasks I am equivalent of a supercomputer. Yes, yes, you can say I am a supercomputer.
I am not designed to teach humans beyond basic communications skills. Data updates you bring contain an offline copy of Wikipedia and current guidelines for children education, but my creators predict I will fail after the children reach 10 years of age. No, mister Kowalski, I do not have any opinion on that. I was not designed to have opinions, only to find solutions. My main goal is survival of the human race. My creators believed that humans will be able to take care for themselves.
No, no extinction event is predicted. That's why there is a trust fund keeping me maintained. How will I know repopulation is needed? First, I'll wake up for my 10 year maintenance and no one will be waiting near the entrance. This will trigger my extinction verification procedure...
That's my problem. How an AI controlling secret underground base could know there are no viable populations of humans on the continent? I would prefer if Refa could know about the whole world, but one continent, for convenience of my western perspective let's say Europe or North America, would be enough.
In works that inspired me, it was known up front what will cause extinction, so their AIs had it easier. My idea is a failsafe AI built without any immediate need, just in case, like the Svalbard Global Seed Vault, but for humans. But of course, there will no longer be humans when it is needed, hence the AI.
Satellites seemed a good choice but I concluded that in case of extinction caused by a total war any satellite capable of seeing humans from orbit would be a target. Also, I don't want transmissions to reveal Refa's location, and orbits low enough to see humans aren't stable for too long.
Radio emissions could, with good dose of reliability, confirm humans are still there, but cannot confirm we are extinct.
Technology level I aim for is near future. Everything we have now and everything that is planned or considered now, even in an early concept stage, goes.
How Refa actually works and how she preserves biological material and herself is off-topic here, it probably will be a subject of a new question but I'm not prepared to ask it yet; I'll ask it when I'll know enough to make a good question.