𝐎𝐩𝐞𝐧 𝐖𝐞𝐢𝐠𝐡𝐭𝐬 𝐀𝐫𝐞 𝐍𝐨𝐭 𝐎𝐩𝐞𝐧 𝐕𝐚𝐥𝐮𝐞𝐬.

This trips up nearly every organisation I speak to.

And it is an expensive mistake.

The reasoning usually goes like this. We will run an open model. On our own infrastructure. Trained further on our own documents.

Then it is ours. Our knowledge. Our standards. Our judgement.

Half of that is right.

A downloaded model does not arrive neutral. It arrives with a specification, a reward target, and the accumulated preferences of the people who shaped which answers were considered better. All of it is already baked in.

You can change the instructions.

You can adapt the model.

You can fine-tune it on your own material.

But none of that automatically gives you sovereignty over its judgement.

Fine-tuning can change how a model behaves. It can improve performance, style, and domain fit. But it does not, by itself, transfer your professional culture: your risk tolerance, your ethical defaults, your escalation habits, your way of deciding what “good” looks like.

So you can end up with your content expressed in someone else’s cognitive grammar.

Here is the distinction worth taking into your next vendor conversation.

𝐂𝐨𝐧𝐟𝐢𝐝𝐞𝐧𝐭𝐢𝐚𝐥𝐢𝐭𝐲
Keeping your data out of somebody else’s pipeline. Open weights solve this well. Buy it.

𝐒𝐨𝐯𝐞𝐫𝐞𝐢𝐠𝐧𝐭𝐲
Keeping your judgement, your professional culture, and your decision rights. Open weights barely touch this on their own.

Most buyers want the second and are paying for the first.

And yet, the thing which protects judgement is not usually a bigger model or another procurement cycle.

It is a boundary.

Not just what the model is allowed to say.

Where it has to stop.

Let AI draft the termination letter. Do not let it decide who is let go.

Let AI draft the pricing options. Do not let it set the price.

Let AI summarise the medical file. Do not let it recommend the treatment.

The model can still do the writing.

Your people must still do the choosing.

Ask vendors for a decision-rights map, not just a deployment diagram.

Because the control point is not the model.

It is the boundary around the decision.

#AIGovernance #AIStrategy #Leadership #AIBias #DataSovereignty

Contact

bruno.gentil@sherpaconsultingasia.com

© 2026. All rights reserved.