Nibras Stiebar-Bang’s Post

𝗔𝗹𝗲𝘅 𝗞𝗮𝗿𝗽 𝗷𝘂𝘀𝘁 𝘀𝗮𝗶𝗱 𝘁𝗵𝗲 𝗾𝘂𝗶𝗲𝘁 𝗽𝗮𝗿𝘁 𝗼𝘂𝘁 𝗹𝗼𝘂𝗱. On CNBC this week, the Palantir Technologies CEO went after the closed AI model business, saying "something has gone completely wrong." His argument: enterprises are burning budget on tokens with little to show for it, and worse, they're handing their most sensitive operational data to labs they don't control. His words: CEOs are afraid they'll "waste my time on tokens" while the labs "get my IP." Technical buyers, he said, want control over their compute, their models, their data stack, and their alpha. They want to own the means of production. Strip away the token complaint and there's a much bigger point underneath it: if your data, your prompts, and your IP are leaving your walls to reach a model, you no longer fully own them. Karp keeps landing on the same word — control — and the only architecture that actually delivers it is self-hosted. 𝗧𝗵𝗮𝘁'𝘀 𝗻𝗼𝘁 𝗮 𝗣𝗮𝗹𝗮𝗻𝘁𝗶𝗿 𝘁𝗮𝗹𝗸𝗶𝗻𝗴 𝗽𝗼𝗶𝗻𝘁. 𝗧𝗵𝗮𝘁'𝘀 𝘁𝗵𝗲 𝗲𝘅𝗮𝗰𝘁 𝗽𝗿𝗼𝗯𝗹𝗲𝗺 𝘄𝗲 𝗯𝘂𝗶𝗹𝘁 HikmaAI 𝘁𝗼 𝘀𝗼𝗹𝘃𝗲 — 𝗮𝗻𝗱 𝘄𝗲 𝗮𝗴𝗿𝗲𝗲 𝘄𝗶𝘁𝗵 𝗵𝗶𝗺 𝗼𝗻 𝘁𝗵𝗲 𝗳𝘂𝗻𝗱𝗮𝗺𝗲𝗻𝘁𝗮𝗹𝘀. HikmaAI is self-hosted, full stop. We deploy inside your infrastructure — your hardware or your private cloud — not ours. Your logs, prompts, and customer data never leave your perimeter, never transit a third-party API, and are never used to train anything outside your walls. You own the data. You own the IP. You own the compute it runs on. Every enterprise racing to deploy autonomous agents is facing the same trust gap Karp described — except it doesn't stop at the model. It's the agents making tool calls, touching customer data, and executing actions with nobody watching. You can pick whichever model you like, but if the enforcement layer watching your agents isn't inside your own perimeter too, you still haven't solved the problem — you've just moved it. Self-hosting the model without self-hosting the security around it is half a fix. That's why HikmaAI's Intelligent Gateway lives on-premise with you. We don't ask your agents nicely to behave — we sit at the network boundary, inside your infrastructure, and enforce it in real time. No SDK inside your agents. No data sent out to prove it worked. Karp is right that the industry has a control problem. We just think the fix isn't a different model — it's refusing to let your data, your IP, or your security controls leave your building in the first place. Sovereignty isn't a model choice. It's a self-hosted architecture choice, top to bottom. Link to article in comments 👇🏼. #AIsecurity #AIsovereignty #AgenticAI #EnterpriseAI #HikmaAI

  • No alternative text description for this image

This point is so valid; “if your data, your prompts, and your IP are leaving your walls to reach a model, you no longer fully own them.” and completely in line with a data sovereignity policy. Didn’t you discuss this as well Jaspar Roos ?

HikmaA intelligent gateway solves this problem allowing enterprises to decide what can go to these APIs either from the autonomous agents running on their platform and on employees desktop. Moving on premise remise solves just one part of the problem. With that said, smallest models are also less resistant which means you need better real time agentic security and HikmaAI gives that.

See more comments

To view or add a comment, sign in

Explore content categories