INPOWERD Leadership Briefing: The Attack Surface Is Now the Substrate of Decision: What Leaders Must Govern Before AI Outpaces Them

INPOWERD Leadership Briefing: The Attack Surface Is Now the Substrate of Decision: What Leaders Must Govern Before AI Outpaces Them

By Earl Shockley, President and CEO, INPOWERD LLC

Trust • Accountability • Service

For most of my 49 years in this industry, I've guarded something physical. A fence. A substation. A control room door. For the last two decades, that shifted to something digital: a network, a credential, a control system. Across five different chairs I've sat in over that career, field operations, control center operations, national-level regulatory, consultant, and now leadership, I've protected reliability the same basic way every time: by hardening the system.

Today, I'm watching this industry begin to guard something else entirely: the decision itself.

That is not a metaphor. It's the argument I'll be making in full on November 11 at the Industrial Security Conference in Copenhagen. It's also worth leaders hearing the short version now, before that room full of security professionals gets the long one.‍

"The Attack Surface Is Now the Substrate of Decision" - Earl W. Shockley

The Shift No One Voted For

Three prior revolutions changed how this industry works, and it's worth being specific about what each one actually did, because the pattern matters.

The Industrial Revolution transformed labor. It moved this industry from human muscle to machine power, steam, then electrification, then the generation and transmission infrastructure that put reliable power behind nearly everything else we'd go on to build. The work didn't disappear. It moved from doing the labor to operating the machines that did it.

The Digital Revolution transformed information. I lived through this one directly. Paper logs, manual meter readings, and voice-only coordination gave way to SCADA, real-time telemetry, email, personal computers, and eventually mobile devices in the control room. The job's purpose didn't change, operators were still accountable for the same reliability outcomes, but the volume and speed of information they had to process changed completely. Those who were slow to adapt to it found themselves behind the eight ball.

The Automation Revolution transformed execution. Protective relaying, automatic generation control, automated switching, and energy management systems meant the system could carry out defined logic on its own: if this condition occurs, take this action, without waiting on a human to execute every step. Operators shifted from executing each action to supervising the systems that executed it for them.

Artificial Intelligence transforms decision-making itself, and that is a fundamentally different kind of shift than the three before it. Automation executes logic a person already defined. AI increasingly helps decide what the logic should be in the first place: which forecast to trust, which risk to prioritize, which anomaly actually matters. That is new territory for this industry, in a way the first three revolutions, for all the disruption they caused, never were.

For the first time, critical infrastructure organizations are beginning to share cognitive responsibility with machines. That is a genuinely new condition, not an incremental one, and it raises a question every leader in this industry needs an answer to now, not someday: what happens to reliability, security, and public trust once machines are participating in the decisions that determine them?

Why the Force Multiplier Cuts Both Ways

I've written before that AI, used correctly, is a force multiplier for judgment, not a substitute for it. Human Intelligence brings judgment, experience, context, and ethics. Artificial Intelligence brings speed, scale, pattern recognition, and analytics. Multiply the two together and you get a real capability gain, not a talking point: Artificial Intelligence amplifies Human Intelligence. Together they become a force multiplier.

But a force multiplier amplifies whatever you point it at, including your blind spots. That same capability gain is exactly why the decision itself has become worth attacking.

Why the Decision Is the Target

For decades, an attacker who wanted to hurt this industry had one real path in: breach the network, steal the credential, reach the control system. We've spent two decades, and a great deal of money, getting harder to breach that way. It's still not a closed door, but it's a far more defended one than it used to be.

That's exactly why the decision layer is the softer target now. An attacker doesn't need to seize a breaker or trip a control system to cause harm. They need the model recommending maintenance priorities, flagging anomalies, or forecasting load to be wrong, and wrong in a direction that benefits them. The system keeps running. The dashboards stay green. The organization simply starts making worse decisions, one plausible-looking recommendation at a time. The mechanics are worth being specific about, because this kind of manipulation doesn't happen at one point. It can happen at three.

1. Before the Decision: The Data Feeding It

The first point of exposure is upstream, in the operational data, external market and weather information, grid and market data, and internal reports that feed a planning model or an operational system before anyone, human or machine, reaches a conclusion. An adversary doesn't need to breach a control system to shape an outcome here. They need to quietly manipulate, corrupt, or inject false data into what's feeding the decision. A manipulated weather feed that understates an approaching load event. A falsified maintenance record that masks a developing asset risk. A quietly poisoned training set that teaches a forecasting model the wrong definition of "normal." None of it looks like an attack. It looks like data.

2. During the Decision: What the System Has Learned to Trust

The second point of exposure is the model itself, the AI, the automation, the pattern recognition that converts that data into a recommendation. AI and digital systems learn from data and patterns, and if the inputs were compromised upstream, the model has no way of knowing it. It develops a trusted but false conclusion, delivered with exactly the same confidence as a correct one. A planning tool recommends deferring a transmission upgrade because the risk model never saw the true loading trend. An anomaly detector stays quiet during an actual intrusion because it was trained to treat that signature as routine. The system isn't malfunctioning. It's doing precisely what it was built to do, on top of a false premise.

3. After the Decision: How It Reaches the Person Who Must Act

The third point of exposure is the last mile, how that compromised insight is delivered through dashboards, alerts, reports, or automated actions, to the person or system responsible for acting on it. This is where human deference does the most damage. An operator trusts a dashboard that reads green, not knowing the summary behind it was quietly corrupted upstream. A compliance report gets generated and distributed before anyone questions the data that produced it. An automated action fires on a recommendation nobody reviewed, because removing that review step was the entire point of automating it.

Three stages. Three different points of failure. And at all three, the organization's real exposure was never the technology. It was whether a human ever had the chance, or the inclination, to ask whether the input, the model, or the output could be wrong.

This is harder to detect than a conventional intrusion, not easier. A breach trips a log, an alarm, an access record. A manipulated decision looks exactly like a normal one, right up until something fails. And it scales in a way physical sabotage never could. Corrupt one model or one data pipeline, and you can influence every decision that touches it across a fleet, a region, or an interconnection, simultaneously, without ever touching a breaker or a firewall.

That's the new terrain. The future cyber battlefield may be the space between human judgment and machine recommendation. The attacker doesn't need your control room anymore. They need your trust in the model that advises it, and the less we question that model, the less they need to work for it.

The Risk Isn't the Algorithm. It's Deference.

Technology replacing human judgment is nothing new, and neither is the risk that comes with it. The Hewlett-Packard HP-35, the world's first scientific handheld calculator, shipped in 1972 with a known flaw: decimal digits would overflow under certain conditions. People trusted the output anyway, because it looked authoritative. That instinct hasn't gone away. AI just gives it a far more convincing interface, and a far larger attack surface behind it.

In high-reliability industries, deference is dangerous, and it is also an opening. When leaders stop asking questions because "the model said so," risk accumulates quietly, whether that model has been compromised or is simply wrong. The obligation to think, question, and decide does not disappear because the math looks convincing. After an outage, a breach, or a compliance failure, "the algorithm decided" will not satisfy a regulator, a board, or a customer. It shouldn't.

The Governance Question Every Leader Must Answer

Every organization adopting AI needs a clear answer to one question: where does AI belong in the decision process? I sort it into three categories.

What can be shared: asset prioritization, maintenance optimization, forecasting, planning scenarios, risk analysis, resource allocation.

What can be delegated: data collection, pattern recognition, monitoring, routine analysis, scenario generation. AI executes; it does not own the outcome.

What can never be delegated: public safety and emergency response, reliability risk acceptance, regulatory accountability, ethical trade-offs, crisis leadership. That category doesn't shrink as the technology improves. It can't. Accountability doesn't transfer to a model.

 "AI is a force multiplier for judgment, not a substitute for judgment" - Earl W. Shockley

The Equation Leaders Actually Own

I describe reliability through a simple equation: Reliability equals Technology times Governance times Culture times Public Trust. All four are required, and three of the four are leadership responsibilities, not technology ones.

The first century of electric reliability was built on physical infrastructure. The second century will be built on cognitive infrastructure. The question in front of every leader in this industry is no longer whether AI will participate in decisions. It's whether you're prepared to govern those decisions and defend them against anyone trying to manipulate them before, during, or after they're made, before the technology outpaces your ability to control it.

If your leadership team hasn't yet answered the three-category question above or asked who in your organization would even notice a model being quietly manipulated before, during, or after a decision, that conversation is worth having before your next AI initiative goes live, not after.

"The future grid will not be defined by the intelligence of machines. It will be defined by the wisdom of the leaders who use them" - Earl W. Shockley

‍

‍