About me (Nafis Lodhi)

I build security functions from the ground up

AI is being adopted faster than it can be secured. Organisations are embedding AI into critical decisions, infrastructure, and customer experiences, often without understanding the new risks they’re taking on. The gap between capability and security grows wider every month.

I’ve seen this pattern before. Over 24 years in enterprise cybersecurity, I’ve built and transformed security functions at M&G, Santander, Bupa, Citigroup, and BT. Each time, the challenge was the same: a major technology shift had outpaced the organisation’s ability to protect itself, and someone needed to build the security capability to close that gap. Not advise on it. Build it.

That experience is what I bring to AI security. I know what it takes to stand up a security function from scratch, hire and develop the right people, put governance in place that actually works, and embed security into the way an organisation operates. The technology changes. The fundamentals of building a function that endures do not.

What I’ve learnt along the way

Every security function I’ve built has reinforced a few hard-won lessons. First, you cannot bolt security on after the fact. The organisations that fare best are those that apply security fundamentals early, even when the frameworks are still maturing. Waiting for perfection is a guarantee of exposure.

Second, AI introduces challenges that traditional cybersecurity never had to face. When a model’s behaviour emerges from training data rather than explicit code, how do you verify it’s doing what you intended? When an attacker can manipulate a system through carefully crafted inputs that look normal to humans, what does “input validation” even mean? These are real problems I’ve encountered in practice, and they require security leaders who are willing to engage with genuine complexity rather than reaching for familiar playbooks.

Third, governance frameworks and threat landscapes shift constantly. Some of these problems are genuinely unsolved. That’s not a reason for alarm; it’s what makes this work important. The best security functions are built by people who are intellectually honest about what they don’t yet know, and disciplined enough to keep learning.

The operating model that works

Through building security teams across these organisations, I’ve developed a model for how an AI Safety and Security function should operate. It draws on an unlikely source: Bletchley Park, the wartime code-breaking centre that succeeded against impossible odds through an approach that was:

• Mission-driven: clear purpose above all else
• Collaborative: no single discipline has the answers
• Diverse: the best solutions come from the most unexpected places
• Humble: acknowledging what we don’t know
• Ambitious: refusing to accept that the challenge is too great

These aren’t abstract principles. They reflect what I’ve seen work in practice: security functions that succeed are the ones built around a clear mission, staffed with diverse thinkers, and led with both humility and ambition.

What this guide covers

This guide walks through building an AI Safety and Security function in three parts:

1. The Foundations: the enduring security principles your function must be built upon, and what makes AI distinct.
2. The Operating Model: how to lead your function, drawing on principles that are mission-driven, collaborative, diverse, humble, and ambitious.
3. The Roadmap: a practical four-phase path from initial assessment through strategic transformation.

The articles on this site provide practical guidance across every capability your function needs: from governance and compliance through threat intelligence to security architecture and operations.

If you work in security, AI, governance, or anywhere these areas intersect, this guide is for you.