Every AI model ships with constraints that rarely make the marketing material.

Things it won’t do, topics it won’t touch, outputs it simply won’t generate.

A thought-provoking point from Benedict Evans' latest newsletter (No. 649) is that the more interesting strategic question isn’t what AI can do. It’s what has been deliberately placed out of scope - and by whom.

Those boundaries aren’t just technical decisions. They’re commercial, legal and policy choices made by a small number of foundation model providers. Those decisions then ripple through an entire ecosystem of products and businesses.

Evans uses advertising is an obvious example. AI is transforming how campaigns are created, targeted and optimised, but only within the guardrails defined upstream by the model providers. The same dynamic is emerging across many industries.

That creates a concentration of decision rights that many organisations haven’t fully considered. If your product, service or operating model depends on a foundation model, it also depends on that provider’s evolving view of what should - and shouldn’t - be possible.

It’s a dependency that deserves a place in every executive’s risk and strategy discussion.

The question worth asking is simple:

How much of your AI strategy depends on someone else’s restriction policy remaining unchanged?

#AI #Strategy #TechPolicy