Sep 10, 2026 · 32 min · 10 segments
This GO AI episode covers what business owners need to know for safely implementing AI, including the security benefits of local AI models, specifically small language models (SLMs). It features an…
Yagub RahimovGuestConnorHostTo dig a bit more into the security benefits of this, could you give us a little more of an indication why they would choose a local small language model? I mean, one of the things that comes to mind immediately for me is sending data to a cloud environment.
I mean, that already has just a lot of, you know, you're putting the data out there in a pipeline that ends up somewhere that you may not even own or have access to.
But, you know, perhaps a cyber criminal may get access to.
I mean, that's for starters.
Are there any other security issues or if there's any subtleties that I missed there that you would like to get into?

So this is actually a very timely question in a way, right? We saw last week OpenAI Hugging Face incident.

Can we go ahead and claim that either OpenAI or Hugging Face did not have security issues? They have, being honest, they have all the technology to their own fingertips.


What happens on the LLM powered security that people are utilizing is that you can actually trick it.

The same thing that is not supposed to be doing something, you can trick it to do exact that one thing.

I'll give you a statistics from today's read up that I was doing on blockchain world the AI powered blockchain transactions enablement environment 62% of the times you can actually convince the agent AI agent to give you the money back 62% of the times, right? And why? Because you can actually prompt inject it and build a different environment.

When you are building and approaching it with SLM side of the activities, that one is not a large language model.

It's a binary discriminator that puts in there that this is social security from the contextual perspective.

And, or this is a context based prompt injection, trying to convince me to send money back.
That's one.
To dig a bit more into the security benefits of this, could you give us a little more of an indication why they would choose a local small language model? I mean, one of the things that comes to mind immediately for me is sending data to a cloud environment.
I mean, that already has just a lot of, you know, you're putting the data out there in a pipeline that ends up somewhere that you may not even own or have access to.
But, you know, perhaps a cyber criminal may get access to.
I mean, that's for starters.
Are there any other security issues or if there's any subtleties that I missed there that you would like to get into?

So this is actually a very timely question in a way, right? We saw last week OpenAI Hugging Face incident.

Can we go ahead and claim that either OpenAI or Hugging Face did not have security issues? They have, being honest, they have all the technology to their own fingertips.


What happens on the LLM powered security that people are utilizing is that you can actually trick it.

The same thing that is not supposed to be doing something, you can trick it to do exact that one thing.

I'll give you a statistics from today's read up that I was doing on blockchain world the AI powered blockchain transactions enablement environment 62% of the times you can actually convince the agent AI agent to give you the money back 62% of the times, right? And why? Because you can actually prompt inject it and build a different environment.

When you are building and approaching it with SLM side of the activities, that one is not a large language model.

It's a binary discriminator that puts in there that this is social security from the contextual perspective.

And, or this is a context based prompt injection, trying to convince me to send money back.
That's one.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.