The AI control gap: Who gets to say ‘It’s safe’?

There’s a simple problem at the center of this argument: AI can look ready before anyone has proved it’s safe to use in the real world. SiliconANGLE AI calls that the control gap — the space between what a system can do in a demo and what the evidence supports trusting it to do in production. Their fix is blunt: don’t let vendors grade their own homework.

That point lands harder because the White House accord cited here only goes so far. It calls for external assessment and board-level oversight, but it does not create a public authority with mandatory access, compliance rules, or penalties for failing to comply. A promise is one thing. A system that can force remediation, restrict use, or shut down a high-risk deployment is another.

Matt Calkins of Appian frames the case in old-school industrial terms: if a sector can do damage, government should require responsible behavior before the damage happens. He compares it to banks holding enough funds and power plants having safety procedures. Waiting for lawsuits after the fact, he argues, is the wrong way to run something this risky. And because the damage from AI is still relatively limited, the cost of fixing the problem now is lower than it will be later.

The China argument gets treated here as a distraction unless it can survive scrutiny. Yes, competition with China matters. But Calkins’ view is that China’s pace still depends on U.S. innovation, and that Beijing has its own reasons to keep AI under Party control. Even if China catches up, the authors say that does not erase the need to prove control before handing over more power to frontier models.

What customers are worried about is much more concrete than geopolitics. The examples pulled from Qualitate’s data talk about agent sprawl, undocumented tools left running after employees leave, and personal or customer data slipping into external AI workflows. That’s a governance problem, a security problem and a privacy problem all at once. The answer, in this telling, is independent experts, public standards and an authority with teeth — plus a business opportunity for vendors that can make oversight less painful.