Don't Give Away Your Edge

Alex Karp, the CEO of Palantir, recently did something on live television that most chief executives would never do. In a blunt CNBC appearance on July 1, 2026, he accused the big AI labs of "stealing the weights and alpha" of the companies that use them. The day before, Palantir had published a nine-point manifesto on "AI sovereignty" with one clear message: enterprises should hold on to control of their own data, their own models, and the infrastructure that runs them. Behind the theatrics is a warning every board should take seriously.

Every company you sit on is being told the same thing. Spend on AI, spend now, spend big. Most of that money flows to a handful of large model providers. Few directors are asking the question that matters most. When your company uses these tools, who actually gets smarter, you or the vendor?

Your company's edge is not its software. It is the things rivals cannot easily copy: your proprietary data, the judgment your best people apply, and the workflows that make you faster or safer than competitors. Finance people call this alpha, the return you earn because you know or do something others do not. AI can capture and compound that edge inside your business, or it can quietly move it to a vendor who sells it back to you and eventually competes with you directly.

That is not hypothetical. Figma built its product deeply around one AI lab's models. Months later, that same lab launched a competing design tool. Even if every contract term was honored, the lesson is clear. Once a provider sees how your business works at scale, the temptation to move into your market becomes hard to resist.

So the real question for a director is not which AI model is best. It is how much of your company's independence you are prepared to hand to someone else.

The manifesto boards should read

Karp's argument rests on one word: sovereignty. Sovereignty is your company's ability to make its own choices about its own future. When you hand a vendor your data, your models, and the systems that run them, you are not just buying a service. You are handing over part of that ability to choose. Later, the vendor can change the terms, raise the price, or use what it has learned from you to compete with you, and you will have little say.

Two of these points deserve translation for a boardroom. "Your data is your treasure" means that every contract, ticket, design, and decision you feed a vendor's tools teaches that vendor how your business wins. And "controlling your weights" matters because a model's weights are simply the settings inside an AI system that hold everything it has learned. When a vendor tunes a model on your work, your hard-won know-how gets baked into weights that sit on their side of the wall, not yours. That is what Karp means by labs "stealing weights."

Why you can't just trust the contract

So how does a company with almost no leverage over the big providers actually keep its sovereignty? Not through tougher negotiation, and not by trusting a promise you cannot verify. You keep it by design. Three moves make sovereignty real, and none of them depend on believing a vendor.

1. Treat the model as a swappable part. Models are becoming interchangeable. Design your systems so you can switch providers without rebuilding anything important. That ability to walk away is the real leverage you have with a vendor.
2. Control what the vendor can see. Instead of letting your people paste contracts and strategy into a vendor's chat product, route AI through your own systems that send the model only the minimum it needs. The vendor returns an answer and never sees the full picture.
3. Keep the valuable part in-house. The models are rented. The record of how your business works, the rules it must follow, and the logic of your workflows should be built and kept in software you own. That is the asset that compounds on your side, not the vendor's.

Your business, written down

Sovereignty is not only about what you keep away from vendors. It is also about what you build for yourself. The most valuable thing to build here is a clear, written record of how your business actually works. Your customers, contracts, suppliers, claims, patients, whatever your core things are, how they connect, and the rules that govern them.

Right now that record is scattered across systems, policy binders, and the heads of a few people you cannot afford to lose. That was tolerable when software was just a tool. It becomes dangerous the moment you let AI act inside your company.

Here is why it matters, in plain terms. A raw AI model sees "an invoice." Your written record lets it see "an invoice from a flagged supplier, on a high-priority project, in a sanctioned region," so it flags the problem instead of confidently approving it. Same model, far better and safer behavior, because the context lives in your systems and not the vendor's.

This is Karp's real point about Palantir. When he says his company's job is to make powerful models "safe and useful and precise" in military, intelligence, clinical, and industrial settings, he is describing exactly this record, the layer where reality and rules are written down so a model can be powerful without being reckless. You do not have to like Palantir, or buy anything from it, to take the point seriously. If that record is yours, every AI interaction makes your understanding of your own business sharper. If it belongs to a vendor, every interaction makes theirs sharper instead.

What to do: a director's checklist

You do not need to become technical. You need to make five things happen, and hold management accountable for each.

1. Put a sovereignty rule in writing. Right now most AI decisions get made one deal at a time, usually by whoever is keenest on a new tool, and usually to save time or money. No one is accountable for protecting the company's independence. A written, board-level rule fixes that. It sets a standard every AI decision has to be tested against, it gives your managers a clear line they can point to when they turn a vendor down, and it gives the board a basis to hold them to it. Keep the rule simple: we will keep control of our data, our models, and the systems that connect AI to our operations, and we will not let any vendor learn from our core work in ways that let them compete with us.
2. Ask for an honest map of what is already leaking. Request a plain inventory. Where does our real business logic live, which AI tools already receive our data and under what terms, and where are AI systems already making decisions? You are looking for places where your crown jewels are already flowing into systems you do not control.
3. Fund a small team that owns this. A handful of people who sit with your operators, capture how the business really works, write it down as your own record and rules, and keep it under your control. Make them accountable to technology, operations, and risk together, not buried inside IT.
4. Fix contracts, but protect the crown jewels with design, not paperwork. Tighten vendor terms on data use, retention, and training. Since you cannot fully verify those terms, do the thing you can control. Keep your most valuable, differentiating work out of any vendor whose terms you do not trust, and use those vendors only for routine tasks.
5. Change the scoreboard. Stop rewarding "we rolled out X copilots," a number that tells you nothing. Ask instead which workflows you improved, what business result followed, what guardrails are in place, and which learning loops you control. Then fund the projects that build advantage you own, and starve the ones that mostly train someone else's model.

The bottom line

Karp's outburst drew attention because it sounded like a rant. It matters because it was not really about his company. It was a crack in the polite consensus that has told boards to spend aggressively on AI in ways that leave them more dependent and less different from everyone else. Sovereignty over your data, your models, and your infrastructure is not a technical detail. It is the difference between a company that owns its future and one that rents it. This is a decision directors cannot hand to a CTO or outsource to a vendor. You can only delay it, and every month you delay, you may be teaching someone else's system a little more about how to run your business. That insistence, that the company keeps control of its own edge, is your job.

Go deeper. Protecting your edge is lesson two of The AI Boardroom Edge, a free six-day course for board directors on governing AI and turning it into value across your boards and portfolio. Enrol free →