Building trust in AI: an operating model approach

Organisations are weaving AI into the fabric of how they work, hoping to capitalise on the technology’s potential to unlock new value and transform their business. But a quiet anxiety is growing alongside the opportunity: “How do we know we can trust what our AI systems are doing - and how do we build that confidence at the pace we want to move?”  

Getting this right requires more than a policy or adopting the most impressive technology. Sustained confidence comes from solid foundations: the right operating model. But many organisations are missing this fundamental step.  

Taking an operating model approach to this problem will enable organisations to move with certainty, and to adapt successfully as technology, regulation and the competitive landscape continue to evolve. 

The AI ‘trust gap’ 

As Agentic AI creates more autonomous systems, the ambition to act and reap the benefits pulls one way, while the confidence to account for behaviour often pulls in the other. These two forces create what is becoming known as the ‘AI trust gap’: the distance between what organisations want AI to deliver and what they can confidently stand behind. 

In practice, the gap surfaces in uncomfortable ways. Regulators are asking tougher questions, customers want to know exactly how their data is used, and leaders are nervous about organisational readiness.  

Potential scenarios include: 

• If a regulator asked for evidence of how an AI system behaved six months ago, what could you provide? 
• How do you know an AI agent has acted within its authorised boundaries - and who is accountable when it has not? 
• When an AI agent takes an action in the world - calling an API, modifying a record, triggering a downstream process - are you confident that action is observable, auditable and recoverable? 
• Do your people have the processes, skills and working practices to manage AI effectively - including the judgement to know when to challenge it, override it, or escalate? 
• How well are you managing cyber risk? Can you withstand new attacks on the vulnerabilities that emerge when powerful tools are adopted faster than security controls can keep up? 

Taking an operating model approach to closing the AI trust gap 

When organisations think about creating trust in AI, the first instinct may be to produce a policy. This matters, but its only part of the picture. Ethics sits at the heart of trust too but no single framework, whether based on duties, outcomes or values, is enough on its own. 

Organisations making real progress are developing their human infrastructure: the ability to navigate uncertainty, use judgement and adapt. The right conditions foster curiosity, sensible decision-making, and a genuine learning mindset, which help teams use AI responsibly, not just compliantly. 

In most organisations, this work is already happening, with legal, risk, compliance, engineering and product teams all working on AI. But joining that work up requires getting the operating model right. In practice, this will mean creating clear accountability, implementing workable governance practices, and enabling decisions that ‘stick’. 

Without that, the AI trust gap widens. Trust fades when ways of working don't make the right thing clear, consistent or observable. 

Building trust as a connected set of AI capabilities 

If we view the AI trust gap as an operating model problem, closing it means being deliberate about which capabilities need to be in place and how they connect. 

Our AI capability map helps to visualise this across two layers: foundational capabilities for harnessing AI tools effectively, and transformational capabilities for driving long-term value. Trust runs through both but concentrates in a handful of areas. 

Part of this is technical. The AI operating system - the architecture, observability tooling, cyber controls and operational practices put in place - determines whether governance intent is enforceable in practice. It is what gives people the visibility to see what AI systems are doing, uncover problems early and intervene when something isn't right. As AI becomes more autonomous, this infrastructure matters more. 

But the technical layer only takes you so far, and this is where many organisations are at risk of underinvesting. Many capabilities needed to close the AI trust gap are deeply human: ethics frameworks that set clear boundaries and guide judgement when lines are unclear; risk management disciplines that connect AI behaviour to the organisation's broader appetite for risk; governance structures that give policies real teeth; and the data practices that ensure models are built on trustworthy foundations.  

Perhaps most underestimated of all is the talent and cultural conditions that allow people to use AI well, such as the capacity to know how AI actually works, good judgement, the ability to navigate ambiguity and real accountability for outcomes. These are not soft considerations sitting alongside the serious work. In most organisations, they could be the biggest drivers of the gap. 

How to make progress on AI trust 

Trust emerges when teams work together under a common direction, translating principles into coordinated action across people, process, information, structure and technology. Here is how organisations can start moving with intention.