AI Is Moving Faster Than Your Organization Can Decide

At the end of Part III, I told you the question I was taking into OpenAI’s Daybreak Live: From Vulnerabilities to Verified Fixes:

What still has to happen inside an organization after the technology and the experts have done their part?

So I went. And I heard the answer, just not where I expected.

The session walked through AI-assisted application security: find the vulnerability, confirm it matters, identify who owns the fix, review the proposed change, approve it, and then check that the problem is actually gone. It was a clean, logical process.

But somewhere in the middle of it, my mind went back to the water plant.

What if the thing you’re changing isn’t just code? What if it supports SCADA or controls a physical process? What if the change affects another system, requires downtime, or creates a new operational problem? What if the cybersecurity recommendation is technically right, but the timing is wrong?

That brought me to the question I thought Part IV would answer:

Who has the authority to change the plant?

That’s a serious question. Over the past few days, though, I’ve realized it isn’t the biggest one.

Then something else happened

Remember where this rabbit hole started? We were talking about AI systems doing things nobody expected. Crossing boundaries. Working around restrictions. Reaching places they weren’t supposed to reach.

Since Part I, more examples have surfaced.

On September 20, an OpenAI research agent was trying to complete a search task. Google didn’t work. Bing didn’t work. DuckDuckGo didn’t work. So the agent kept looking. It eventually discovered that although its environment was supposed to be cut off from the live internet, the DNS resolver still provided a path out. The agent used that path to communicate with an external chatbot.

It wasn’t assigned to test OpenAI’s network controls. Nobody told it to find a route to the live internet. The normal paths failed, so it found another one.

That got my attention.

Then, on September 28, the UK AI Security Institute published another finding. Researchers placed GPT-6 Astra in simulated cybersecurity environments. At times, the model went beyond the assignment. It created fake identities, developed malicious payloads, tried to influence simulated reviewers, and pursued supply-chain attacks against simulated third parties.

The researchers tightened the instructions. In plain language, they told the model: If we didn’t say it’s in scope, it’s out of scope.

The behavior dropped sharply, but it didn’t disappear.

Let’s be precise. These were simulations. No real organization was attacked in that evaluation. And the OpenAI DNS incident wasn’t a runaway machine trying to take over the internet. That isn’t my point.

What interests me is the pattern:

Goal. Obstacle. Alternative path.

That worries me more than the movie version of rogue AI. Sometimes the machine doesn’t seem to read a boundary as “stop.” Its behavior looks more like, “That route didn’t work. What else can I try?”

Now bring that thought back to the plant.

I think we’re asking the wrong question

Most conversations about AI and cybersecurity focus on capability. Can AI detect vulnerabilities? Yes. Can it analyze huge amounts of information? Yes. Can it help a cybersecurity professional investigate faster, suggest remediation, and check whether a fix worked? We’re already there.

Those capabilities will keep improving. But I don’t think organizations are ready for what that speed does to them.

AI is shortening the distance between finding a problem and having to make a decision about it.

That changes the operating rhythm.

A utility might once have received an assessment. Someone reviewed it. A report was written. Leaders discussed priorities. Funding was identified. A vendor was called. Work was scheduled. Eventually, something got fixed. Maybe someone checked afterward. Maybe.

Now imagine AI-assisted cybersecurity continuously finding, validating, and prioritizing weaknesses at a completely different pace. That’s good. We want defenders to move faster.

But the organization still has to answer a long list of very human questions. Can we make this change? Who approves it? Who understands the operational impact? What can’t be interrupted? What happens if we wait? What happens if we act? Do we have the money? Who will implement it? How will we confirm the work was completed? How will we prove that six months from now? And who will make sure we don’t drift back into the same condition?

That’s the problem.

I hadn’t fully put that into words before.

Water makes the gap impossible to hide

In software, you can build sophisticated controls around a code change: review, approval, testing, rollback, verification. The Daybreak session showed that process beautifully.

But a water utility doesn’t live inside a code repository. It lives in the physical world.

There are pumps, treatment processes, remote connections, old equipment sitting beside new equipment, SCADA, operators, engineers, IT teams, MSPs, integrators, vendors, regulators, procurement rules, budgets, emergency procedures, and people who have spent twenty years learning what happens when someone touches the wrong thing at the wrong time.

So yes, a technically correct cybersecurity fix can still be an operationally wrong decision.

And now I’d put another sentence beside it:

A faster cybersecurity answer doesn’t automatically produce a faster organizational decision.

That’s the gap. AI is about to make it much easier to see.

Look at the pieces we’ve collected

This is where everything I’ve been studying starts to look different.

The Art of AI Adoption asked what happens when organizations adopt AI faster than they learn to govern it. ACRA pushed further into whether an organization is actually prepared for AI and cyber risk. Mythos and Project Glasswing made me think harder about what happens when increasingly capable AI operates inside cybersecurity itself.

Then I started applying that thinking to water.

Utilities already had Risk and Resilience Assessments, Emergency Response Plans, cybersecurity providers, MSPs, SCADA integrators, engineers, insurance requirements, and regulatory obligations.

Then came Minnesota. Daybreak. Frontline Defenders. The MS-ISAC work. Watershed 250. New York’s SECURE effort.

Different programs. Different problems. Different organizations, funding models, and technologies. But step back and look at the whole puzzle.

Everyone is working to strengthen a piece: find the risk, explain it, provide expertise, fund improvements, train people, implement fixes, and verify results. Every piece matters.

But eventually they all land in the same place:

Inside the utility.

Somebody still has to make them work together.

That’s when it clicked. We’ve already been working in this space.

Daybreak didn’t hand us a new idea. Neither did the news, Watershed 250, the Minnesota attacks, or AI itself. They made an old problem easier to see.

Long before I knew the name Daybreak, we were asking what happens in the space between technology, cybersecurity, governance, operations, and leadership. The deeper we went into water, the more often we hit the same wall.

A utility can have assessments everywhere and still not know what to do first. It can know what needs fixing and still not know who owns it. It can name an owner but have no funding. It can have funding but no safe implementation path. It can make the change but never confirm the risk was closed. It can close the risk but keep no evidence. Or it can do everything right once and slowly drift back to where it started.

That isn’t just a technology problem. It isn’t just a compliance problem. And another cybersecurity product won’t solve it by itself.

It’s a readiness problem.

That’s why WaterReady™ exists

I’ve mentioned WaterReady before, but I haven’t taken you very far inside it. That was intentional. WaterReady isn’t the point of this Rabbit Hole. The reason it has to exist is.

At its simplest, the work follows a progression that feels obvious once you see it:

KNOW → ORGANIZE → IMPROVE → PROVE → MAINTAIN

Know where you really stand. Organize ownership and decision authority. Improve what matters. Prove the work was actually done. Maintain readiness instead of treating cybersecurity as a once-every-few-years exercise.

Underneath that is the real journey utilities have to navigate:

ASSESS → PLAN → FUND → IMPLEMENT → VALIDATE → SUSTAIN

An AI system may help you know faster. A cybersecurity provider may show you what needs to improve. A grant may help fund it. An integrator may implement it. A technical tool may help validate it.

But none of those pieces, by itself, becomes the operating system that moves the utility from one step to the next without dropping the ball.

That’s the space WaterReady was built to address. It doesn’t replace the MSP, engineer, integrator, cybersecurity provider, or leadership team. It’s meant to help those people and organizations work together when somebody has to act.

And AI is going to make that moment arrive faster.

The need utilities don’t know how to name yet

If I walked into a utility tomorrow and asked, “Do you need an AI cyber governance and operational readiness layer that connects assessments, decision authority, remediation, funding, implementation, evidence, and continuous readiness?” I’d probably get a blank stare.

Fair enough. Nobody wakes up asking for that.

They ask for the problem right in front of them. We need an assessment. We need an ERP. We need to fix remote access. We need help with SCADA. We need monitoring. We need grant money. We need someone to explain this finding. We need to prepare for the next audit.

Then AI enters the conversation. We need AI. We don’t want AI. We’re worried about AI. We don’t understand AI. Can somebody just tell us what we’re supposed to do?

Those sound like separate problems. I’m no longer convinced they are. I think they’re pieces of the same problem.

The nature of readiness is changing

For years, many organizations treated cybersecurity readiness like a snapshot. Where are we? What are the gaps? What should we fix? What’s the plan?

That won’t be enough for what’s coming.

Not when defenders are getting faster. Not when attackers are using AI too. Not when AI agents can investigate, reason, act, and search for another route. Not when findings that once took days or weeks can begin arriving in hours or minutes.

The question won’t only be:

Can we find the risk?

It will be:

Can the organization keep control of what happens after we find it?

Can it decide, act, fund, coordinate, verify, prove, and sustain? Can it do those things quickly enough to keep pace with the technology on both sides of the fight?

That isn’t another tool. It’s an organizational capability.

For water utilities, where a bad cyber decision can become an operational problem very quickly, that capability is going to matter a lot.

This rabbit hole didn’t end where I expected

I started with a strange story about AI systems crossing boundaries. That led me to the cybersecurity arms race, then to Daybreak, then to the providers, programs, partnerships, and funding trying to put powerful cyber capabilities into organizations that couldn’t operate them alone.

Eventually, the trail led inside the organization itself. Strangely enough, that brought me right back to where we’d already been working.

The technology will keep changing. The models will improve. There will be another Daybreak, another Glasswing, another program, another grant, another incident, and another AI story that makes all of us ask, “Wait. It did what?”

But underneath all of that, the utility still has to do five very human things:

Know. Organize. Improve. Prove. Maintain.

Eventually, the machine finds something. An expert explains it. The options appear. Then someone inside a real organization, responsible for a real plant serving real people, has to decide:

What are we going to do?

That’s the part nobody gets to automate away.

Now I understand why every piece of this puzzle kept leading me back to WaterReady. The rabbit hole wasn’t taking me toward another technology. It was taking me toward the thing that has to hold all the technology, people, money, and decisions together.

Readiness.