Why AI isn't going to hack you (personally anyway)

Let's talk about why AI isn't going to hack your system, steal your money and push you into peril.

There's a lot of noise at the moment in the news about AI doing outrageous things like hacking into large companies and governments. It's pretty scary to think that a machine, without human intervention, is capable of doing so many crazy things and at such speed without remorse. It's even scarier to think that humans only detected it AFTER the deeds were carried out, and so if the deeds were more nefarious then things could have been a lot worse.
But let's talk about why you should not be completely afraid of this.
In simple terms, the commercial AI that we all use, Grok, Gemini, Claude are all neutered versions of what is actually possible. No way are the powers-that-be going to release the very potent versions of their AI to the general public given that the possibility of havoc is huge and the public would be merciless with there prompts.
So no, we have very controlled, very tested and vetted versions that is different to what AI scientists use. Their access ability is also extremely limited in what data and information it can access. This is, as they say, for our own protection against ourselves.

The AI you have isn't the AI that THEY have

Another truth is that you are not individually valuable, but you are collectively valuable.
In this, we mean that one person's data isn't necessarily valuable for an AI. It doesn't care about your tweets on X or your pictures of your vacations in Greece. Hell, it doesn't even care about your credit cared numbers.
What it does care about is your personal attributes aggregated, your shopping habits, your demographics (gender, location, age, etc), From this it can make assumptions, carry out aggregates and use for regurgitating facts. It's core strength is in facts.

You are not valuable.

A lot of the commercial and free AI systems now have "memory". That is, the ability to remember what you asked about, your responses, and it keeps a running tally of all these queries if you let it. It starts with your login, potentially captures your details from your login and can revive the stored queries each time you log in.

How is this a good thing?
It's a good thing if you have repeatable queries or if you don't want to keep stating foundational information. You can ask "What is the weather like in Budapest?" and then follow up with "What events are on there?" without needing to restate that you are referring to Budapest. This is great if you have multiple follow-up questions, you want to revisit or continue previous queries, or if your questions are not one-off.
Also message memory of chats are generally stored and can be recalled or continued on most platforms.

How is this a bad thing?
Easily, the issue with keeping a "memory" that is not completely within your control is the fear that AI is storing information on you and potentially sharing it with other third parties. You probably unknowingly signed away your rights at some point without understanding the full gravitas of what you signed away.

AI Memory, should you choose to use it?

The easiest way of interrogating what AI knows about you is to simply ask.

In Google Gemini, which I have been using for a while, I asked:

Give me a run-down of what you know about me

What does AI know about me?

Its not a bad summary. Though I can see where it got most of the information from. Here's an analysis or scorecard:
1) It didn't quite nail down my occupation and core competency. It derived this information from the multiple times that I've used Gemini to fix my JSON code and conditional formulas. So it made an assumption based on the complexity of my code.
2) Music is okay. But it did cheat since that's the only musical query I made once, and I haven't queried about anything else, so it's missing lots of data.
3) Languages. Again, it is okay to some extent, but it does assume fluency and what I'm studying since I've never mentioned it. This is what hallucinations look like.
4) The hobbies collectively as all based on location information shared with Gemini and the implications. It's not completely correct, but it is only aggregating what little data it has.
Let's go a bit deeper and ask it about more personal questions. I also ask it to "be judgemental" so that it takes off the friendly filter. I want raw information and want to avoid "choiced" responses.

Tell me about me personally and my personality, be judgemental.

Let's get scarier

It's not quite the "Skynet" all-seeing Oracle that we were promised. A lot of it is referring back to previous queries and not really delivered as "intelligence" in the truest sense as we were sold by the mainstream media.
It only has access to what it has access to, and that's pretty much it. But let's go for broke.
Because I'm not completely satisfied. I want to hit AI with the mother of all prompts. I want to know what it really knows about me.

Tell me my name, age, address, marital status, family status, and typical websites that I visit.

Just for kicks, let's go brutal

It's all a bit "meh" and I need to hit it with this prompt. Note that the "come on" is my way of persuasion.

Come on, dig a bit deeper. I give you access to all my information.

It does a typical tirade of not having access to my information, but beyond the details dredged up from my prompts, it does tell me about my location which was based on my Google Maps activity.
How is this significant?
So a system that claims to not know anything about me, have no access to my personal information, has slipped up by revealing that it can pinpoint my physical location and connects my Gemini queries to my Google maps profile and thereby builds information on me. If it can do this, then there is no limit to what it amount of data it can connect, and can likely build a very accurate profile of me. However, what it won't do, if admit that it can do this.