AI Is About to Get a Lot Less Helpful

Right now, you can ask an AI about network security, then about a legal loophole, then about how to reverse-engineer a competitor’s pricing model — and it will answer all three without blinking. This feels normal already. It shouldn’t. It’s strange, and I think it’s temporary.

Not All Knowledge Costs the Same to Give Away

There’s a gap we keep glossing over. A tip about cooking or fixing a bug in your code — giving that away costs nothing. If a million people learn how to make a better risotto, the world is slightly nicer and no one is worse off.

But a financial strategy, a security exploit, a frontier research method that compresses years of trial and error into one answer — that’s different. That’s closer to capital than conversation. It’s the kind of knowledge that moves money, accelerates discovery, or shifts competitive advantage. It was expensive to produce. It sits inside these models precisely because someone paid for it.

Today, you can ask an AI to walk you through a tax strategy that sits in a legal grey area, or explain a drug interaction not yet in the public literature — and it will answer. That’s the open phase. The question is how long the people who built and paid for that knowledge will keep letting it flow out for free.

AI Will Become a Gatekeeper, Not an Assistant

The direction is clear: AI systems will become more controllable, and more restrictive. Not clumsily restrictive the way today’s keyword filters work — intelligently restrictive. They will read the intent behind a question, not just the words. Two people asking the exact same thing will get different answers, because the system will have read different goals, different permissions, different intentions behind the request.

What AI answers freely today will simply be off-limits tomorrow.

Take two people asking the same question about a software vulnerability — a student studying security and a professional on a paid engagement. Today they get the same answer. Tomorrow, the system reads who they are, what they’ve subscribed to, what their account signals about their intent — and one gets the full picture while the other gets a vague overview.

This will be called safety, and some of it genuinely will be. But safety and control use the same machinery. The same system that blocks a dangerous instruction can block anything its owner prefers to sell, license, or keep scarce. Once you build the gate, the question is no longer technical — it’s political: who decides what goes through, and for whom?

It’s Already Happening

The clearest signal doesn’t come from a prediction — it comes from one of the labs building these systems.

Anthropic has kept some of its most capable models out of public reach entirely, accessible only internally or through select partners, while publicly available versions remained more constrained. Now they are releasing Fable — a model explicitly designed to be more controlled, with deliberate restrictions on specific domains and types of questions.

Read that again: a frontier AI lab is intentionally building a model that refuses more. Not because it lacks the capability — because it has been decided that certain knowledge should not flow freely. The answer exists inside the model. What’s being gated is the permission to receive it.

This is not a safety patch. It’s a deliberate choice about who gets what, dressed in the language of responsibility. And Anthropic is not alone — every major lab is moving in this direction at different speeds and with different justifications.

The gates are not coming. They are being installed right now.

Power Is Shifting From Owning Resources to Owning Knowledge

For most of modern history, power meant controlling physical assets — land, energy, supply chains, money. Those who controlled the inputs controlled what could be built. AI adds a new layer on top of that. Whoever controls the most capable models gains something more precise: direct control over knowledge itself. Not just over what people can do, but over what they can understand.

A model that sits between billions of people and how they learn, decide, invest, and create can shape behavior at a scale nothing has ever had before. It can be tuned topic by topic, user by user. It can be open here and quietly closed there. It can help some people move forward while gently steering others away — one by one, at scale, invisibly.

The same small set of actors who already control finance and infrastructure are on track to also become the gatekeepers of knowledge. That is a new and serious concentration of power, and it doesn’t require any conspiracy — just ordinary incentives.

We Are Still in the Open Phase

The broad, generous AI we use today is the starting point, not the endpoint. It’s the phase before the gates are built — before the lawyers, the licensing deals, and the competitive pressure make staying open a liability.

Picture what this looks like in five years: the same AI that today explains any topic to anyone may require a medical credential to discuss clinical research, a paid license to run financial modeling, a verified affiliation to touch anything near sensitive science. The knowledge is still there — inside the model, trained on it, shaped by it. It’s just no longer yours by default.

Whether the more controlled AI to come serves the many or the few depends on decisions being made right now, while the rules are still being written and the systems are still open. That window won’t stay open by default.