Artificial intelligence is offering us an extraordinary bargain.

AI is already collapsing the economics of knowledge work. Months are becoming weeks. Teams are becoming individuals. Projects that once required millions of dollars can be accomplished for a fraction of the cost.

The exact numbers will differ, but the direction is clear. AI is making intelligence cheaper, faster and scalable in ways that were unimaginable only a few years ago.

Businesses cannot ignore that opportunity. Neither can governments or individuals.

But bargains have two sides.

What if the apparent price of frontier AI isn’t the real price? What if we are also paying with our information, intellectual property and accumulated knowledge?

More people are beginning to ask that question.

What Are We Giving Up?

Millions of people now share remarkably personal information with AI. We ask it about relationships, finances, careers and problems we might once have discussed only with people we trusted.

Businesses are doing their own version of the same thing.

They are connecting AI to proprietary information accumulated over decades. Governments are experimenting with the technology across institutions holding information about their citizens, operations and decision-making.

The immediate attraction is obvious: The more context we give AI, the more useful it becomes.

But that is precisely what makes the bargain worth examining.

Palantir CEO Alex Karp has been unusually blunt about this. He argues that companies risk transferring their “alpha” into AI systems they do not control.

Alpha is more than data. It is what an organization has learned about how to compete.

Companies may be transferring the very thing that makes them valuable.

Karp goes further. He argues that cheap tokens obscure the true economics of the exchange. In his telling, customers aren’t simply buying inexpensive intelligence. Their information can make the AI system more valuable too.

His provocative question is the right one:

“What is the true cost? Not just what you’re paying.”

That, to me, is the AI grand bargain.

The real cost may not be the token bill. It may be the value you surrender.

The IP Question Is Already Here

The fight over AI and intellectual property is no longer theoretical.

The New York Times and other publishers are suing OpenAI and Microsoft. Ziff Davis has sued OpenAI. Major music publishers have pursued Anthropic. Authors have brought large-scale copyright claims, including litigation that resulted in Anthropic’s $1.5 billion settlement. More recently, Apple sued OpenAI, alleging the misappropriation of trade secrets as OpenAI expands into hardware.

The claims differ, are contested, and many remain unresolved. But their growing number points to a much larger economic question.

Companies have spent decades and enormous sums creating intellectual property and accumulating knowledge. Now the courts are being asked to determine when AI companies can use that information to build increasingly valuable products of their own.

Every business should be paying attention.

What is truly preventing a frontier lab from eventually replicating important parts of your business?

Contracts and enterprise protections matter. But are terms written by the company on the other side of the transaction, and accepted with a click, really an adequate long-term strategy for protecting the knowledge that makes your company valuable?

That isn’t an accusation of misconduct. It is a question about risk, incentives and the extraordinary value now attached to information.

Has the Market Priced the Liability?

There is another dimension to the grand bargain: Liability.

Frontier AI companies increasingly warn that more capable systems could create catastrophic risks. Those warnings deserve to be taken seriously.

But they also create an unusual economic contradiction.

Investors are assigning enormous valuations to companies whose leaders simultaneously warn that future versions of their technology could cause extraordinary harm. At the same time, more conventional liabilities are already emerging through intellectual-property litigation.

Karp takes the argument further. If the potential harm from frontier AI is truly as large as some of its developers warn, the resulting liability could eventually exceed what any private company could realistically bear. His argument is that government may ultimately be asked to limit or assume some of it.

That makes the industry’s growing calls for AI regulation worth examining from another perspective.

Some legal scholars and policymakers argue that much of the necessary framework already exists. AI does not operate outside the law. Criminal law still applies to criminal conduct. Civil liability, including negligence, product liability and other established causes of action, can already apply when AI causes harm.

So what precisely must new AI regulation accomplish?

There are legitimate arguments for new rules, particularly where existing law was never designed for increasingly autonomous systems. But regulation also determines liability. Depending on how it is written, it can establish standards of care, create safe harbours or other protections, and shift responsibility among developers, deployers, users and governments.

It can also shape competition. The largest frontier companies can absorb compliance costs that smaller companies and open-model developers cannot.

None of this means the warnings about AI safety are manufactured. But incentives matter.

Governments should therefore examine calls for AI regulation through more than the lens of safety.

Who bears the liability? Who receives protection from it? And what kind of AI market do those choices create?

Enter With Your Eyes Wide Open

None of this is an argument against AI.

I believe the opposite.

The grand bargain may prove to be one of the greatest productivity opportunities businesses, governments and individuals have ever encountered.

But we should enter it with our eyes wide open.

And we should understand that there are alternatives to simply transferring our accumulated knowledge into someone else’s model.

My thesis is that the future of enterprise technology will increasingly separate an organization’s intelligence from the models used to reason over it.

Organizations can maintain control of their own knowledge and application layers while allowing different AI models to reason over them. Models can improve or be replaced without taking the organization’s accumulated intelligence with them.

In that architecture, the model is extraordinarily important.

It just doesn’t own the relationship.

Canada’s Version of the Bargain

Canada needs to have the same conversation.

The federal government is already embracing AI. The Canada Revenue Agency, for example, says its employees have access to Microsoft Copilot and that its approved GenAI tools can support work up to and including Protected B.

There may be sound technical and contractual protections around those deployments.

But who is responsible for scenario-planning the larger sovereignty question?

Microsoft is an American company. U.S. law, including the CLOUD Act, can compel providers subject to U.S. jurisdiction to produce data within their possession, custody or control through lawful process, even when the data is stored outside the United States.

That does not give the U.S. government unrestricted access to Canadian government information.

But it should force us to think more deeply about what sovereignty actually means.

A Canadian data centre is not necessarily the same thing as Canadian control.

Canada does not need to build its own version of every frontier model. It should use the best technology available.

But it also needs Canadian-controlled infrastructure and home-grown technology because dependency creates strategic vulnerability.

Sovereignty is ultimately about choice.

Can You Walk Away?

Perhaps the simplest test of the AI grand bargain is this:

Can you walk away?

If a better model emerges, can you switch without losing what your organization has learned?

If circumstances change, can sensitive work move somewhere you control?

If the answer is no, adoption has become dependency.

The AI grand bargain is too valuable to reject.

But its true price is more complicated than the invoice.

Use the best intelligence available. Understand what you are giving in return. Protect the knowledge that makes you valuable.

And preserve the ability to walk away.