Editor’s note: This is part one of a two-part blog series. The second installment will publish next week.
It’s 11:40 p.m. on a Sunday.
Someone on your team has just spotted something odd in an AI tool’s output. Maybe a model is quietly leaking data it should never have touched. Maybe three departments have been using a shadow AI application nobody approved. Maybe the output simply feels wrong, and they cannot yet explain why.
They are staring at their phone, thumb hovering over your number, running the calculation every experienced operator knows:
Is this worth waking someone up for, or will I look like I panicked over nothing?
That 15 seconds of hesitation is your risk culture.
Not the policy binder. Not the dashboard. Not the beautifully colored governance framework nobody can remember without opening SharePoint.
That moment, repeated a thousand times a year across an organization, determines whether you hear about trouble while it is still a weak signal or after it has acquired a project code, external counsel and its own steering committee.
AI has not changed that fundamental truth.
It has made it harder to ignore.
Gartner’s C-level Communities 2026 Leadership Perspective Survey in March 2026 across 1,085 CISOs found that enabling and protecting AI had become the number one priority for security leaders. Fifty-six percent cited the pace of change as their biggest challenge, while 52 percent said they did not yet have the skills required to deal with it.
That is not a confession of incompetence.
It is evidence that the job has changed shape faster than the training for it has.
And the answer is not simply more AI.
The world got weirder, and it is not going back
We use VUCAD to describe the volatile, uncertain, complex, ambiguous, and digitized organizational risk environment.
These are not academic labels. They describe Tuesday.
Volatile means conditions change before the quarter ends. A model update lands. A regulation shifts. A competitor launches an AI capability nobody expected until next year.
Uncertain means the organization may have plenty of data and still not know what happens next. A confident AI answer can be confidently wrong.
Complex means the risk rarely belongs to one system or one team. It runs through vendors, cloud platforms, business processes, data pipelines, employees and customers.
Ambiguous means reasonable people can examine the same evidence and reach different conclusions. Was that outcome a mistake, misconduct, an unintended model behavior, or something nobody realized had become automated?
And digitized means more of the organization’s activity, judgement and exposure now depends on software, data, automation and AI.
That is the environment CISOs and CROs are operating inside.
AI is part of the answer to VUCAD. It is also part of the problem.
It can process evidence at extraordinary speed. It can identify anomalies, correlate weak signals, summarize mountains of information and model possible outcomes.
But AI cannot decide what your organization should tolerate.
It cannot tell a powerful executive that the assumptions behind a much-loved program are weak.
It cannot create psychological safety.
It cannot decide who should own an uncomfortable consequence.
It cannot make somebody pick up the phone at 11:40 p.m.
AI already does some of this. It triages the alerts a junior analyst once triaged, and it drafts the assessment a human now only signs. That sharpens the problem rather than settling it, because the work being displaced is exactly where people used to acquire the judgement we are asking them to supply later.
AI can improve the signal. Culture determines what happens to the signal next.
That is why the human side of the CISO and CRO role is becoming more important, not less.
1. Judgment: knowing what matters when everything can be measured
Digital organizations do not suffer from a shortage of information.
They suffer from a shortage of attention.
Add AI and the abundance becomes almost comic. More alerts. More dashboards. More summaries. More predictions. More red dots telling senior executives that something somewhere deserves immediate attention.
When answers become cheaper, judgement becomes more valuable.
The CISO and CRO increasingly need to distinguish signal from noise, possibility from materiality, probability from consequence, and model confidence from business confidence.
Judgement runs before a proposal exists. It decides what deserves attention at all, which is a different job from testing a plan somebody has already written. It requires questions no machine can finally answer for you:
What matters here? What assumption drives this conclusion? What evidence would change our mind? What happens if the model is wrong? What decision are we actually being asked to make?
This is where experience matters. So does mentoring.
Someone facing their first serious AI incident makes a different decision if they can call somebody who has already sat through the 2 a.m. version of it. Good mentors do not hand over answers. They transfer pattern recognition.
They let somebody else benefit from scars they did not have to acquire personally.
One part of judgement gets discussed least because it is the least flattering. Judgement degrades under fatigue, ego and the wish to look unbothered. The 11:40 p.m. call is made by somebody who has already had a long week and would rather not be wrong in front of you. Knowing your own failure mode under pressure is a capability, not a personality trait, and it is the one most likely to decide the outcome.
Takeaway: Do not ask only whether your people understand AI. Ask who helps them think when the framework runs out.
2. Translation: making risk mean the same thing in every room
AI does not respect your org chart.
A single AI deployment can create technical, operational, legal, financial, conduct and strategic risk at the same time.
Yet many organizations still discuss the same issue in several rooms using several dialects.
Security sees an attack path.
Risk sees exposure against appetite.
Technology sees architecture.
Legal sees liability.
Finance sees cost.
The business sees a deadline.
None is necessarily wrong. The problem begins when nobody translates.
The CISO who says, “We have 37 critical vulnerabilities across externally exposed infrastructure” may be technically precise and still leave the executive team unsure what to do.
The useful translation sounds more like:
“Three systems supporting customer transactions remain exposed to a credible attack path. We can accept the interruption risk for another quarter, or bring remediation forward now.”
One is a finding.
The other is a decision.
The job is not to simplify complexity until it becomes inaccurate. It is to simplify it until somebody can act.
That means getting CISOs, CROs and business leaders into one decision conversation, rather than three sequential briefings in which everyone politely assumes somebody else connected the dots.
Takeaway: If cyber, risk and the business describe the same AI issue differently, do not create another slide. Create a common consequence.
3. Challenge: testing the decision before reality does
VUCAD environments reward speed.
AI adds another complication. Machine-generated answers often arrive with the appearance of authority.
A number has been calculated. A recommendation has been produced. The model sounds confident.
And humans are remarkably good at mistaking precision for truth.
That makes constructive challenge one of the most important skills a CISO or CRO can develop.
Challenge means being willing to ask:
What would have to be true for this plan to fail? What is absent from the data? Which risk are we consciously accepting? Is this deadline real, or merely convenient? What would cause us to stop?
It is especially difficult when the program has a powerful sponsor, the economics look attractive and everyone wants to move.
Good challenge is not obstruction.
The objective is not to win the argument. It is to improve the decision before reality conducts a considerably more expensive review.
Courage without judgment becomes noise. Judgment without courage becomes silence.
Takeaway: For every material AI decision, require somebody to argue the strongest credible case against it before approval.
Part I gets the risk onto the table and tests the quality of the thinking. That is only half the problem.
In our upcoming Part II, the question moves from the individual to the organization. Will people surface weak signals? Does somebody own the call? And when the world changes, will yesterday’s decision change with it?
