You can use AI in OSINT. Just audit it.

"You can't use AI in OSINT." I understand where it comes from. Models make things up. They're confident when they're wrong. And an investigation that rests on something the AI invented isn't an investigation.

All true. But the conclusion doesn't follow. You can use AI in OSINT. You just have to know its limits and be able to check its work, the same way you'd check a junior analyst's.

How we wired it in

In Ubikron, AI is optional. The product is completely usable without it, and a lot of our users never switch it on. When you do:

• It writes the fluff. Descriptions of clippings, drafts of the text around your findings. The tedious part.
• It works from your case. When you ask it something, it answers from the pages, notes and graph in your project.
• It leaves a trail. If the AI browses (in Claude for Chrome, say), every page it visits is captured in Ubikron like yours. Every enrichment it runs is logged. You can see exactly where it went.
• Its graph work waits for you. Nodes the AI creates stay grey until you give them a color, and the links it draws are dotted. That's by design: nothing it adds looks like your work until you've looked at it and approved it.
• It never writes in your report. It has its own section. You take what you want from it, or nothing.

That last point matters most. The investigator stays the one who decides what's true. The AI just gets you there with less typing.

Don't trust it. Audit it. Then decide.

Roelof TemminghFounder, Ubikron