Think back and try to name some of the defining moments of 2022. There are more than a few to pick from. It was the year that saw Russia begin its full-scale invasion of Ukraine, the year Queen Elizabeth died at 96, and the year Argentina took home the last World Cup.

At the risk of offending Argentinians everywhere, however, probably more important than any of those events was the fairly quiet launch of a new kind of software.

Since the introduction of ChatGPT in November 2022, marking the point when real-world AI first crept into the popular consciousness, evidence shows the technology has seen a staggering uptake, more rapid than PCs, the internet, and even the Tamagotchi.   

Yet AI is still something of a black box, the equivalent of a digital Ouija board, taking in our questions, along with oceans of data, and spitting out its verdict with eerie certainty, though most of its users never know the how or the why behind its thinking.

Even the developers of some of the most popular AI models, like OpenAI and its closest rival Anthropic, admit they don’t understand how their creations work, while a recent study from tech giant Apple has argued that AI reasoning is nothing but an illusion.  

None of that has stopped millions of us from quickly trusting these systems, even with our most important, sensitive decisions. 

It’s a shift most obvious in the business world, where the vast, rapid rollout of AI has led to a FOMO moment for leaders. Recent IBM research found that as many as 64% of CEOs acknowledge the risk of falling behind is driving investment in technologies they don’t have a clear understanding of or business case for. 

AI is not only being deployed to make gains in productivity, but to make decisions about who should be hired, who should be fired, what products should be pursued or dropped, and how best to defend operations against hostile attacks.

That’s a lot of trust to put in systems which, for the most part, are opaque in their judgments. And that’s the other factor separating AI from all those other great leaps of technology: the measure of its success depends on our belief in its powers.

But there’s a problem. 

AI’s Trust Deficit – and What it Means for Business

Earlier this year, KPMG published the results of a huge survey into the issue of trust in AI, taking in the insights of 48,000 people across forty-seven countries.

The findings suggest that while most workers already use AI regularly and expect benefits from it, fewer than half of the general public trust the technology. In fact, trust levels have fallen since 2022, while worries about AI systems have shot up from 49% to 62% last year.  

“It’s definitely an interesting stat,” says Leanne Allen, a partner in KPMG’s financial services tech consulting practice and head of AI advisory for the consultancy giant’s UK arm.

“The trust challenge is coming from a lack of transparency around what’s being done with AI, coupled with the concerns around the bad use of AI. That’s fuelling distrust, because it’s flooding us with the potential harms and the risks.”

This lack of confidence is having real-world repercussions, as the majority of global businesses scale back their AI investments specifically because of trust issues, while KPMG’s survey found trust gaps are also manifesting internally, with workers choosing to hide their AI use or use it in ways that contravene policy, like uploading sensitive information to public AI tools. 

“The research suggests people are concerned about the ethics and the fairness of AI,” says Allen.

“It also translates to how organisations are thinking about this. Most are focusing on corporate services and back office efficiencies rather than really pushing the boundaries on how they can use AI, so there’s a level of risk and concern that they’re not yet comfortable with.”

However, this cautious approach may be risking a golden opportunity.

A McKinsey study from May suggests that businesses proactively implementing responsible AI practices are seeing major benefits over peers taking a ‘wait and see’ approach.

Companies that have invested in aspects like explainability, fairness, and transparency have seen cost reductions (42%), increased consumer confidence (34%), enhanced brand reputation (29%), and fewer AI incidents (22%).

“For a B2B or B2C business, you’ve got to build trust or clients won’t engage,” warns Allen, “and if you’re going to move towards externally facing, for example, with smart virtual assistants, customers aren’t going to want to engage without trust either, so you’re wasting your spend and you’re not going to see ROI.”

But as some organisations begin pulling ahead, many firms have been left struggling to operationalise responsible AI at scale.

Part of the problem is technical. Just 22% of enterprises report their IT infrastructure is ready for AI, and only around one in 10 rate their data as fit for AI decision making.

Then there’s the human element. Workforce resistance driven by widespread fears around future job loss is undermining companies’ efforts to implement responsible systems.

“The risk I’m seeing is, leaders think they just need to worry about the upfront training, and that’s not the case,” says Allen.

“The reason is you’re changing a behaviour, breaking old habits to form new ones. Your entire workforce has to have a level of understanding and confidence, and the ability to work differently with these tools.”

Better training could go a long way toward easing worker tensions, but while 60% of companies claim to offer training, fewer than half of workers have received any. That makes it harder to embrace, leaving AI as a necessary, obscure evil that they need to use or risk being left behind.

“Organisations need to be very mindful about explaining to the workforce what their future role would look like in an AI-enabled world.

“They have to bring people on that journey.”

What Effective AI Governance Looks Like

The real question is, how do you make trust tangible enough for employees to buy in? By definition, it’s an ethereal concept, aspirational rather than operational.

For Allen, who has helped guide multiple KPMG clients through these murky waters, the key is to break trust down into actionable components. 

“The way we look at trust is thinking about different stakeholder groups,” she explains.

“So there’s organisational trust, employee trust, then there’s customer or broader societal trust, and you’ve got to look at regulatory trust as well.”

By taking a granular, bottom-up approach, Allen argues that companies can create AI systems that deliver real accountability and foster confidence for everyone involved. 

Take organisational trust, for instance, which essentially comes down to ensuring AI systems are clear and understandable, making sure datasets are safe and secure, and having the right checks in place to keep tabs on how the tech is being used.

“Organisational trust can be tackled through things like transparency, explainability, ensuring a focus on privacy angles, cybersecurity risks, and putting in appropriate controls around those aspects to make sure there’s accountability over what is and isn’t being done with AI across the organisation,” says Allen.

Multinationals like Google, SAP, and Microsoft are taking this approach, having created frameworks that aim to establish these kinds of principles as the bedrock of future AI. 

Recommended reading

• PwC Interview | How Agentic AI is Reshaping Business
• Microsoft and PwC Publish AI Playbook for Businesses
• How to Avoid Wasting Months on AI That Doesn’t Work

And it’s not just tech firms either. The H&M Group, for example, has a framework for developing responsible AI based on nine core principles, including transparency, collaboration, and respect for human agency.

That last one is perhaps the most important, as any long-term tech investment or AI strategy hinges on building internal trust among employees, but Allen argues there is a simple solution here, too.

“When you think about the workforce’s level of trust, then the majority of that has got to be focused on the upskilling and reskilling on how to use these tools.”

The unprecedented scale of AI’s rollout, coupled with the severe lack of training that studies like KPMG’s highlight, has led to distrust of the technology. Without the right training, workers are left AI illiterate, unable to trust something they don’t understand. 

“If organisations focus on how to upskill and reskill the workforce to adopt these systems appropriately and more accurately, then they’re going to build that level of trust.” 

Regulatory trust, meanwhile, shouldn’t just be a box-ticking exercise. Companies that freely demonstrate their AI practices align with legal obligations, sector-specific standards, and ethical expectations will have an easier time proving compliance.  

“If you take a trusted by design approach that includes ethics, you’re typically going to be compliant with any regulations that come along because you’ve thought about all those potential risks and issues along the way, and you’ve been mitigating them,” Allen explains. 

“If you can provide evidence through transparency and explainability, then the regulators are going to be pretty pleased with you.”

In the end, though, successful governance isn’t only about building trust; it should be rooted in intent. Allen emphasises that organisations need to see AI not as a separate beast, but as a core driver of strategic change.

“You have to have a clear purpose of what you’re trying to achieve,” says Allen.

“AI is not and shouldn’t be treated as purely a tech strategy. It’s a business change strategy at its heart, and that means finding out how AI enables business strategy effectively.