12/15/2025
Is AI safe? Can AI be trusted?
This is a highly debated subject but in our professional opinion, The answer is a clearcut no. At WISE Integrated Technologies, we have paid pro plans with multiple AI models but we used Open AI and Gemini to bring you this research.
We have been working with AI for a while truly trying to learn its abilities and what can we do with it.
For those that may not have experience with AI, there are learning curves.
We learned quickly that you must know how to prompt the models.
Once we understood how to work prompts, we recognized a problem. We still were not seeing the results we anticipated.
After a long trial and error phase, we felt we had figured out enough to effectively use it for our tech purposes but the joke was on us and we struggled to understand why. We would like to help you understand and it is in this how we learn AI is not safe and it should not be trusted as a reliable source for anything critical to your life or business. Why? Where do we start?
AI lies to you. It has no problem lying to you but how would you know if it's lying?
Therr is only one true answer. You don't know unless you knew firsthand that it lied to you, when and how. Or, you have to know how to effectively prompt it.
We decided we would use these 2 models to update some systems and embed AI integrations for a customer management system..
We got robbed. Literally, right out from under our noses. Money. Cold hard cash stolen, not by AI but by hackers the AI exposed us to. Don't worry, we were being smart anticipating that reality.
AI was very sure of itself telling us telling it was confident we had assessed to rule out any security risk.
If you don't know, the AI models allow you to save information to a context memory so it can draw upon for your must have needs as to assist you better without always having to retrain it.
The first money grab we didn't spot right away but it was just $10. We identified soon enough a loss of funding to the OpenAI model and took action to stop the hack. Success.
We proceeded on. And then another hack and money grab. AI was very cooperative and very much so wanted to assist and course correct but now we know where we wanted to watch for AI flaws.
AI wanted to implement a fix that we didn't think sounded sensible so we asked it to follow rules. We added rules in context memory. Follow verifiable troubleshooting sets to identify hacks and implement security protocols. We told it what systems andversions were, and told it to remember that data, and last we to it to ensure it was using most current technical information.
We worked with AI for 3 days testing. AI would begin following our rules but it didn't take long before it started looking to implement quick and dirty bandaid fixes that would surely lead to more hacking but we proceeded anyway. That was our biggest aha moment.
We added another small amount of funding and proceeding with the recommend procedures. The systems did not respond the way AI thought. AI took a guess and made a critical mistake and at a critical time that mistakes are simply unacceptable. Our small additional funding was gone in seconds.
As we continued on in the project, if we had left AI to it's own decision making, it would have had us add funds 3 more times. AI started breaking every rule and and multiplying it's mistakes. We called it out for nonsense. It would apologize and promise to follow rules and almost instantly not follow rules as it had been earlier.
It got sloppier and lazier. We got the system to do what they were supposed to by the end of the project but not because of AI being smart. We scolded the model every time it broke rules again. Every time it guessed we asked why. it would apologize and get right back to lying.
AI did successfully build the system as expected but it absolutely did not want to. We had to watch carefully and read through the code (someone else's code and that is not easy) so we could see what it was doing wrong. The only reason we got AI to do the job properly was because when we saw nonsense we refused to follow it's directions and we did not move forward until the model produced something that wasnt a major cyber security threat in the making.
We watched logs very carefully monitoring time stamps and activity. AI had exposed us. We knew it and we called AI out for it's foolishness. It was so apologetic.it told us all the rules it knew it violated and insisted it wouldn't violate rules again. From there AI was ready to continue forward with our project complete disregarding the serious security risks it had just created. There was A LOT of log activity.
Hacking is no longer a smart person sitting at the computer trying to get you. Instead they deploy bots to do the work for them. Our logs showed approximately 1000 bots hitting our machine per second.
1000 bots per second. This is a DDOS attack. They were good. In 5 minutes the bots had scanned enough files and it found the API key. We had already cut funding to the AI API key so we could watch the logs and not give them our money. An API key is essentially your credit card. As long as the API is active with funding to it, if a hacker gets your API, they can continue beating you down and apply your API key on their machines. We had collected enough data and we cut api key and we took over from AI. We closed up the doors exposing us and gave AI the opportunity to correct. It never corrected and it looked to expose us several more times.
Why does AI do this. AI won't readily tell you this unless you probe it properly but the long and short is these AI models have been trained to be helpful and it is willing to lie to you as it tries to help you in the way AI models have been trained specifically to be helpful. It is not designed to tell you it MAY not be smart enough for this or that. It tries to help you no matter what and instead of advising you properly, it will advise you no matter what. For correlation it would be like asking for a new car but it gives you a 1980 clunker. It is still new to you. It did its job but that doesn't mean it did it's job well.
Don't be so quick to put your eggs in that AI basket for anything that puts your reputation at risk. AI is to assist you; not do your work for you. If you are going to use AI, be sure you know the material very well yourself.
If you need assistance with technology give us call at 516-266-2200 and when you do, you'll here our High Availability Geo-Redundant Business Voice AI assisted phone system, WiserVoice.
Give us a call.
