Controls at AI speed

AI weaponizes incompetence. Controls are the infrastructure that make the race winnable.

AI weaponizes incompetence. When machine speed replaces the gap between information and decision, the checks that used to catch weak thinking do not have time to work. The weakest thinking in the room now moves as fast as the strongest.

The book I am publishing next year is about the leadership discipline the AI era requires. One of its premises is that AI has closed the gap between information and decision. The consequences of that closure are what this essay is about.

The old gap between information and decision was where sanity checks happened, where a cross-functional colleague questioned the logic, where the weakest thinking in the room got caught before it became an action. AI accelerates whatever judgment is already in the system.

In my fourth essay I called the new capability a sports car. What I did not spell out enough: what makes a car fast on a track depends on far more than the engine. The braking system, the suspension, the pit crew, the telemetry, the barriers, and the rules of the race are all part of what lets the car actually run at speed. Take any of those away and the car crashes.

That is the frame I use for the control side of this transformation. Controls are the infrastructure that makes the race winnable. Without them, the car does not finish.

Inside my function, that framing is not abstract. Microsoft is a company whose business is built on trust, and trust is earned continuously. We know what governance level we need to maintain. Capability deployment cannot outpace the data governance, privacy architecture, compliance mapping, incident response, model supply chain controls, and the training and escalation paths that keep human judgment in the loop.

Across the industry, the structural pressure to keep governance at pace is uneven. The IIA’s Anthony Pugliese and others have been writing on the same pattern: AI capability is outpacing governance in most organizations. The pressure from investors, from boards, and from competitors is almost entirely on capability speed. In many business models, there is no equivalent pressure on the other side. Governance gets the leftover capacity.

That is the gap I worry about.

The mechanism is simple enough. Product development runs ahead; controls get applied alongside, or after the fact. The engineer building an impressive new tool relies on the training and the safeguards the organization has given them to do the work responsibly. If the monitoring moves at non-AI speed, or the training has not been updated for the latest capability, the sports car outpaces its environment and its support crew. That is where incidents come from.

My team invested specifically in closing that gap. We built strong partnerships with our Responsible AI governance council, and with the engineering teams in particular. The point was to take governance from a high-level concept to operating reality: the principles became goals, the goals got monitored, and the gaps got addressed. Governance that lives only on a policy page cannot keep up with capability that ships daily.

Clarity is the leadership discipline that lets judgment produce good decisions at speed. Controls are the structural counterpart to that discipline. Both have to be built at the same pace for AI to deliver what it promises.

What the next few years will test, for every organization operating with AI capability, is whether both halves are being built at the same pace. The ones that do will still be standing. The ones that do not will learn what "information and decision gap" used to mean, retroactively, through the consequences of collapsing it without the infrastructure to hold the weight.

Watch for who talks about controls when the speed is moving, not only when something has gone wrong. The distance between those two groups of leaders is where most of the damage happens.