The Predictability Factor is a weekly deep dive at the intersection of AI, Security, Privacy and Tech, to help you Go From Chaos to Resilience in The World of AI.

This part 1 of 2 series on AI Governance and Security Maturity and Roadmap. Part 2 coming out soon.

While I was still in Norway, in one of my CISO roles for critical infrastructure, I got the opportunity to work on the National AI strategy, specifically on the topics of AI Security, AI Ethics and Privacy for building Trustworthy AI.

In another CISO role, my team and I implemented our enterprise AI Strategy and Governance, together with our AI and Emerging Tech policy before the first public version of ChatGPT came out.

I’ve been working with ML/AI dedicatedly for the last 6 years or so. While more and more organisations are waking up to understanding the importance of an AI strategy, governance pillars, controls and engineering, I still see so many of them struggle with building and implementing an effective AI strategy and governance pillars within their enterprise.

Every time I talk to CISOs, CIOs, risk leaders, GRC, internal auditors or boards, this is the one question that I get often:

You can’t. At least not entirely. But that’s only half the story.

Just because most security practitioners will say “zero-trust”, the answer is not that simple. That first question is almost always followed by this statement:

That’s the reality of business.

I get it there is no hundred percent trust. There will never be. At the same time, the concept of trust vs. zero trust aren’t opposites, and it falls apart without giving some context. That context is key.

AI agents will go rogue, hallucinate, make completely inaccurate decisions and take completely wrong actions leading to severe unintended consequences.

In this 2-part series, a special edition of The Predictability Factor, we will go through 5-steps and dimensions of AI governance and security maturity roadmap for agentic AI in your enterprise. These are key for building trust, predictability and resilience in your agentic AI enterprise.

Here’s a quick snapshot of my AI governance and security maturity roadmap, along with the keynote I gave recently in Finland where I touched upon my entire maturity roadmap. However, the devil is in the details (step-by-step roadmap below)…

The devil is in the details. In part 1 (this edition of The Predictability Factor), we will cover:

In part 2 (coming out soon), we will cover:

Step 3: Culture and Literacy, Step 4: Governance, Policy and Risk Management, Step 5: Controls and Engineering, and more (along with a bonus).

Partner

Millions of people use Wispr Flow to give AI tools richer context by voice. 89% of messages sent with zero edits. Speak your prompts, skip the typing. Free on Mac, Windows, and iPhone. Try Wispr Flow free.

Over the last 3 years, there has been a major shift in AI. Since ChatGPT’s first public model came out, we have gone from “AI that says things” to “AI that does things”.

OpenClaw is the clearest window into that shift.

The reason I say that is because while you may not have it on your roadmap, some employee in your organisation has already installed AI agents and is running agentic AI in your enterprise within your infrastructure, as Shadow AI.

Why does this matter?

In under three months, OpenClaw accumulated over 200,000 GitHub stars, making it one of the fastest-growing open-source projects on record. 10s of thousands are running it locally on machines connected to your enterprise, with persistent memory across sessions, with access to your company data, integrated with your apps and tools, and operating with real system access. One developer in Austria. Four months. A tool now running inside thousands of enterprises whose security teams have no idea it is there. This isn't just a case of OpenClaw.

AI agents are everywhere. Rather it’d be more accurate to say:

PromptArmor researchers documented a specific attack pathway: send a malicious link to an OpenClaw agent through Telegram or Discord. When the agent responds and includes an attacker-controlled URL, the messaging app's link preview automatically renders it, transmitting the user's confidential data to the attacker's domain. No click required. The preview fires. The data leaves.

341 malicious skills were discovered in ClawHub, the official OpenClaw skill registry, primarily delivering malware designed to steal credentials and data from macOS systems. Gartner labeled OpenClaw "insecure by default." China banned it from government computers.

Just over the last 30 days, we have seen 3 major supply chain attacks: LiteLLM, Axios, and Vercel. Add to that, Mythos, the non-public version was hacked by unauthorised users, by breaching its vendor and guessing the URL.

That’s the state of “Hacking AI” in 2026. It’s not attacking the underlying models. It’s attacking everything around it. Or in this case, just “guessing” the URL of Mythos.

No sophisticated malware. No zero-day. No insider access.

So how do you build “trusted” agentic AI within your enterprise? Let me share an analogy which is key to building the right foundation.

I am an Indian-German, living in London. Over the last 10 years or so, I’ve shared this personal story that somehow so neatly ties to every evolution of technology we have seen thus far. Before it was the Internet, then it was the web, cloud, OT, IoT, IIoT and now it's agentic AI.

I’ve always loved driving. When I drove in Germany, I fell in love with it even more. My friends joke about this. If you know how to drive in India and if you know how to drive in Germany, you likely can drive anywhere, in any condition, well.

German highways are the only highways in the world where you have unlimited speed zones.

My top speed on a German highway has been 240 kph (ca. 150 mph). Now, that’s not normal on any other highway in the world. I love that I can go that fast legally. If you’ve ever driven that fast on the German highway, at that speed, you know 3 things are true for sure. 

But that's exactly what we are doing with agentic AI.

Driving without speed limit works on the German highway because 1) the car is engineered for it, 2) the brakes are built for it, and 3) the road infrastructure is designed with longer curves for high-speed turns.

Remove any one of those, and the speed that made the drive exhilarating becomes the thing that kills you.

Add to that, in the real-world, LLMs function as probabilistic machines providing probabilistic outcomes.

Ask it to follow a “security” rule 10 times, it will comply 7 times. Maybe 8. Maybe 5. That's not a guardrail. That's a wish-list.

Even if you had brakes in your car, imagine one where the brakes only worked accurately 7 out of 10 times. Would you drive it? You wouldn't.

When I drive 240 kph (or 150 mph), on a German highway, I trust the engineering of the vehicle and the physical infrastructure of the road. I trust that when things are about to go out of control, I can use the brakes reliably and predictably, as long as I maintain the safe distance.

The car alone isn’t enough. The infrastructure within and around it matters. I wouldn’t go that speed in the U.K. where I now drive every day. Not just because it’s illegal. But also because the infrastructure around it doesn’t support it, reliably (even if it were legal).

German highways are built specifically with longer curves to allow drivers to turn safely at massive speeds. Add to that stringent and more rigorous licensing, and stricter traffic rules (tailgating is punished, lane discipline is strictly followed, etc).

That brings me to the maturity roadmap for agentic AI governance and security in enterprises.

The above examples illustrate this perfectly. AI adoption is everywhere. Trust is not.

Firstly, how do you even define resilience?

Most people define resilience as something that happens in the moment of crisis or adversity. I see it differently.

Scaling without the right governance, architecture and implementation is like running faster in the wrong direction without brakes.

It won’t get you to the right destination, no matter the speed.

I’ve worked with some of the biggest organisations across finance, healthcare and other critical infrastructure companies across EMEA. The one constant theme I see is:

My 5-step framework is all those learnings of years put together in this one roadmap that I recommend for every organisation that wants to go beyond just an AI pilot, safely, securely, and reliably, while building trusted AI.

What this roadmap is and what it isn’t:

All of the following steps or dimensions of maturity are important. You may decide to invest more in one or the other depending on your maturity, but I would not recommend skipping any one of them.

Let’s dig in.

Early 2023, in my role as the Group Chief Security Officer, my team and I were working together with other key stakeholders in the organisation to roll out an enterprise AI tool for the entire organisation. Data governance was the first key step in that AI rollout. In fact, I provide that example in my practical AI strategy template.

Most people misunderstand why data readiness and validation is key.

Unless and until you have the right data quality, data classification and data governance in place, you are just exposing your data to AI, just waiting for a data breach to happen.

This could be because your AI now has access to data it shouldn’t or just because you didn’t put in the right DLP measures in place and your employees, knowingly or unknowingly have exposed your sensitive data to the Internet. Most of that exposed data gets leaked to the dark web, where the criminals sell your data for profit, espionage or further targeted attacks against your infrastructure.

As we rolled-out the enterprise AI, we had defined 3 key pre-requisites and outcomes:

Here is the part nobody wants to sit with: 75% of large enterprises already have shadow AI running across their networks. 86% are blind to where their AI data is actually flowing.

In 2014, Amazon built an AI hiring tool to screen engineering candidates at speed. By 2015, the company realised the model was not reviewing resumes in a gender-neutral way.

It hated women. Well not literally, but it learned that from the dataset that did.

It was penalising resumes containing the word "women", downgrading candidates from all-women colleges, and favouring action verbs that male applicants used more frequently. The system had been trained on ten years of resumes from a male-dominated industry. Amazon scrapped it in 2018. Rightly so.

This is not a historical footnote. It is the blueprint for what happens every time an organisation gives AI access to unclassified, unvalidated and biased data, but somehow expects sound decisions to come out the other end.

Data readiness is not about clean data. It is about three things working together: classified, unbiased, and access-controlled.

The moment you give an AI model access to your entire data environment without classifying it first, you hand it access to everything your employees were never supposed to see. The identity and access management framework your team spent years building collapses in a single rollout. Employees who should not have seen confidential data now do. Not because the AI was malicious, but because you did not tell it what it should not touch.

And when training data carries historical bias, as it almost always does, the AI amplifies that bias in every decision it makes. At speed. Without pause.

In one of my CISO roles, I was working with my team on a national AI strategy. This was later released and became a key part of the national AI initiative with security, privacy, ethics and resilience as underlying the backbone. Learning from that I implemented AI strategy for organisations in critical infrastructure space. Here’s the biggest pattern I’ve noticed.

Most organisations I work with cannot answer a simple question: where in your business are AI systems currently influencing decisions that affect your business operations, your clients, your regulatory standing or your business outcomes? What exact use cases have you defined that are key for AI to implement and work with? What are the key governance pillars AI will be deployed and implemented on and are they aligned with the overall business objectives?

Last year, I wrote about AI in decision making. A strong AI strategy doesn’t just ask where you plan to use AI, but also where it is operating right now and which business decisions are AI augmenting / replacing.

Especially the ones you don’t know about.

That gap is the problem. Agentic AI is already embedded in M&A due diligence workflows, credit risk modelling, supplier evaluation, and strategic planning tools at enterprise scale. It is not a future scenario. It is today's reality. According to Deloitte's 2026 State of AI in the Enterprise report, while AI agents outpace the necessary deterministic guardrails, only one in five companies has a mature model for governance of autonomous AI agents.

The reason is almost never the model. It is the governance pillars being an afterthought and added after the deployment, if at all.

This needs to start with a strong (yet simple) AI strategy, tied with your AI governance pillars that underpins every AI implementation in your organisation.

"We're using AI because everyone else is", the one sentence I've heard in more board and executive meetings than I care to count. It is not an AI strategy. It is a liability statement looking for a context.

Most organisations have AI adoption plans. Almost none have an AI strategy with governance pillars directly attached to it.

Ask your leadership team to answer these questions right now:

If the answers are vague, or the room goes quiet, your AI strategy “just because others are doing it” doesn’t hold.

A real AI strategy must define business cases with measurable outcomes tied to specific decisions, not to general efficiency goals. "Cut costs" is a starting point and not enough.

The NIST AI Risk Management Framework, released in 2023, gives you four functions to structure from: Govern, Map, Measure, and Manage. But none of them start without a sound strategy. ISO/IEC 42001 provides the management system standard. Both are available. But very few have operationalised it.

As my team and I started the AI implementation project, now more than 3 years ago, the first piece of work was alignment on AI strategy. Not 20 pages long. Just 2-3. Here’s a real-world practical AI strategy template I have used to get started.

Strategy without governance and implementation is just paperwork. And paperwork never stopped a breach.

We will go deeper into agentic AI governance, controls and engineering in the next steps of the roadmap.

Part 2 of this series coming out soon, where I’ll cover the rest of the maturity model and roadmap, and more, along with a bonus for you.