Brad Smith chooses his words with a lawyer’s precision, and when he talks about “rules nobody can read,” he isn’t exaggerating. From the stage of the AI for Good Global Summit, Microsoft’s president dismantled the paradox gripping Washington: the United States is effectively regulating artificial intelligence, but through prescriptions so vague and uncoordinated that they’re indecipherable for those who must apply them. No benchmarks or VRAM specs here; the problem is subtler and structural. Regulatory uncertainty, Smith argued, isn’t a bureaucratic annoyance – it’s a brake on the entire industry.

The core issue isn’t the absence of rules, but their transparency. Competing agencies, non-binding guidance, executive orders that sketch boundaries without defining operational standards: for a company developing or adopting large-scale language models, navigating this mosaic is like designing a building without knowing the seismic code. It can be done, but at what cost and with what risk? Smith’s complaint comes as Europe finalizes its AI Act complete with risk categories and clear obligations, making the contrast even starker.

For AI-RADAR readers, the immediate effect of this regulatory fog is visible in deployment choices. When norms are opaque, organizations handling sensitive data – in healthcare, finance, public administration – find it rational to lock things down. On-premise and self-hosted setups become a way to regain control that compliance doesn’t offer: if the legal perimeter is uncertain, at least the physical one can be governed. It’s no coincidence that more and more organizations are evaluating local stacks, where models run on their own GPUs and data never leaves the company data center. Regulatory uncertainty, in this sense, acts as a surreptitious accelerator of technological sovereignty.

There’s also a competitive asymmetry that gets little attention. Large cloud platforms can absorb the cost of ambiguity with legal teams and lobbying resources; for mid-sized companies that want to fine-tune an open-source LLM without ending up in a maze of interpretation, the message is clear: better keep everything in-house. Uncertainty doesn’t hit everyone equally, and this threatens to widen the gap between those who can afford to bet on the most favorable interpretation and those who must protect against the worst-case scenario.

Smith doesn’t propose a solution, but his intervention signals a fracture: the private sector isn’t asking for deregulation, it’s asking for predictability. In a market where computing power and model quality are racing ahead, the absence of regulatory tracks isn’t freedom – it’s a hidden cost. Balance sheets will reflect it in TCO, delayed projects, and defensive architectures. Those with long memories recall how confusion over rules slowed cloud adoption in regulated sectors for at least a decade. With AI, history risks repeating itself, but with a twist: development cycles are so fast that waiting means being left out. The tangible result is that the most forward-thinking companies are already rethinking deployment pipelines, treating on-premise not as a tactical option but as a default architecture pending a readable framework. And they’re right: in a regulatory void, control is the only certainty.