Keith Shaw: So it was before that we hired private investigators to chase someone or made a case for our house. It then moved to the 2000s when the data broker industry began to boom.
Now, instead of learning advanced Google search techniques or “Google Dorking,” anyone can ask the generated AI model to do the work. Chris Wingfield: That's right. And it takes just 6-7 minutes.
Let AI handle it and it comes with a basic profile or report. It's enough for threat actors to pivot physically quickly from the digital. Keith: What else is AI doing now that we couldn't do before?
So much of this data was already on the internet, but now it feels like AI is making everything faster. Chris: Yeah, the core data was always there. The difference is automation and accessibility.
Just ask the AI to “find this person” and it starts pulling through the context. Also, the rise of autonomous agents – assign tasks, start to see systems running in the background, and take small tools to carry out larger goals.
Keith: Is it like assigning your team to a microassistant? Chris: Exactly. Think of Langflow or Agent AI. Often there are “middle managers” agents who coordinate others. However, effectiveness actually depends on how each model is trained and how it presents personally identifiable information, or PII. Take Claude for example.
They are trained using a method known as “Constitutional AI” to justify whether the request is good or bad. ChatGpt uses reinforcement learning using human feedback, so it's more binary: good or bad, period. After that, there is Grok from Xai. This is much more open and uncensored.
That premise is, “If it's on the Internet, you should be able to see it.” Keith: So, all models handle these privacy and ethical decisions differently. Chris: Yes, and the way you encourage them is really important. You can't say “Find Keith Show.” You must provide a context.
Our research team once got a model for writing malware. Rather than asking directly, by explaining and sysadmin scenarios and stimulating them in stages. Learning how to encourage them properly will help you target yourself more difficult and better understand your vulnerabilities.
Keith: So, most experts probably use multiple AI systems (ChatGpt, Claude, Grok) at once? Chris: 100%. Some platforms, such as Perplexity.ai, allow you to choose the model to use for each query.
You can say “Use Grok for this” or “Use Claude.” Each one excels in a variety of tasks – advanced inference, arithmetic, coding, etc. Keith: But in these search scenarios, accuracy is really important.
I don't just want the results. You need the right one, especially if you're looking for a specific person. Chris: Exactly. For example, if you go to a podcast, it might encourage a model. “I'm about to talk to Keith Shaw.
Here are some public links I found – can you build me a full profile? “This will allow actual power to work from what is called a “pivot point.”
AI handles boring foundations so you can dive deeper and faster. Instead of spending an hour, I check everything for five minutes on violation forums, public records, etc.
