In a recent CNBC Squawk Box interview, Palantir CEO Alex Karp said something has gone completely wrong with how AI is being sold to companies. Coming from Karp, that's not a throwaway line. When he says the current model is broken, it's worth asking what he means: enterprises are being asked to hand over their data to AI labs by default, with no real accounting for what that costs them down the line.
Daniel Miessler summed up the core warning in his newsletter, Unsupervised Learning No. 535:
"He's basically saying if you blindly trust your company to big AI labs they will end up eating the entire business. They'll own everything. The compute, the models, your data, and because of that they'll understand your company and your IP better than anyone at the company."
Hand your data to a frontier lab by default, and you're not just buying intelligence — you're handing over the raw material that makes your company your company. Your customer relationships, your internal know-how, your product roadmap, all flowing into someone else's model.
It's Not Just the Big AI Labs - It's Your Entire Vendor Stack
Here's the part that doesn't make headlines: the AI labs are just the most visible version of a much older problem. Every SaaS platform sitting in your stack today — your CRM, your helpdesk, your ticketing system — is a potential collection point for the same kind of exposure. We've written about this before in SendSafely: Your Sovereign Data Infrastructure: by the time customer data reaches your AI, it's often already passed through a half-dozen vendors who may store it, index it, or quietly use it to train their own products.
Sovereignty isn't just about which model you run. It's about every hop your data takes before it gets there.
Sovereign AI Requires Sovereign Data
This is the piece missing from most "sovereign AI" conversations. You can run your models on your own infrastructure — AWS Bedrock, a private cluster, whatever you choose — and still leak everything that matters if the data feeding those models was never protected in the first place. A sovereign model fed pre-exposed data isn't sovereign. It's just a different leak point.
That's the gap SendSafely Halo closes. Halo sits between your customers and your AI ecosystem as an encrypted trust layer, so sensitive files and data are protected the moment they're submitted — before any vendor, agent, or AI model in your stack ever touches them in the clear.
And this doesn't mean ripping out the tools you already run on. SendSafely is designed to be the trusted backbone underneath your existing process flow, not a replacement for it. You keep using Slack, Jira, Zendesk, Salesforce, Intercom, and whatever else your teams live in day to day — SendSafely has native integrations for all of them, plus a full API and MCP server for anything custom. The difference is that sensitive content no longer has to live inside those platforms to move through them. A reference travels through Slack or Jira; the actual data stays encrypted in SendSafely, visible only to who — or what — you explicitly grant access to.
Trust, But Verify Cryptographically
Here's the uncomfortable truth underneath all of this: you can't fully trust a vendor's promises about your data, no matter how good their intentions are. Policies change, incentives change, companies get acquired.
The only vendors worth trusting with your most sensitive data are the ones with no interest in it at all — and who can prove, cryptographically, that they have no access to it, rather than just asserting it in a privacy policy.
That's the model SendSafely is built on: client-side, end-to-end encryption where SendSafely itself cannot see the content passing through it. Not "we promise not to look." Structurally can't.
Karp's warning is really an infrastructure problem in disguise: if you don't control the pipeline your data travels through, you don't control your AI, no matter whose logo is on the model. Sovereign data first. Sovereign AI second. You can't have one without the other.
Want to see what a genuinely sovereign data pipeline looks like for your AI stack?
Request a demo of SendSafely Halo and find out how to protect your customer data from every vendor in between.