Building Trust in AI: Overcoming Data Privacy Concerns and Legacy Skepticism in Established Companies

Recently I had a long conversation with Ben Hoyle about a topic that keeps CEOs up at night, wakes up compliance teams sweating, and makes curious engineers grin, trust in AI. The headline problem is simple, and messy, trust gets broken faster than anyone can write a policy, and older companies carry scars from data leaks and unclear practices, newer companies often treat AI like a shiny toy. Both reactions are understandable, but neither helps anyone build something that actually works.

Why trust matters, and why it was broken

Ben put it succinctly, “how do you build trust in a system?” That question is a lot more than philosophical, it’s operational. Established companies remember incidents where sensitive data ended up in public training sets, and that memory has hardened into skepticism. Ben described the situation as companies knowing “it’s happening” but not being able to acknowledge it, creating what he called a “black market” use of tools. People were copying and pasting sensitive content into public chatbots because they needed to get work done, and legal, security, and leadership were left to play catch up.

Contrast that with a startup that springs up today. They look at the available enterprise-grade plans, the data protection promised by large providers, and see no obvious reason not to use generative models. They lack the history of breaches, so their risk perception is different. For legacy firms the challenge then becomes how to get back to a state where employees, and leaders, feel safe using AI in productive ways.

A practical route to regaining trust

There’s no silver bullet, but there are practical steps that actually move the needle. Ben and I talked through what worked at a large, 45,000 person company, and the recipe is simple, in theory, messy in practice.

• Secure the board, then empower the grassroots, Ben told me, “you need to convince your board of directors that this is a technology you need to be ready for”, and at the same time give employees tools and communities that let them learn by doing.
• Build a cross functional team, bring together IT, security, legal, cloud, and AI folks, and treat this as a shared problem rather than a siloed one. Ben described that moment after a high profile leak, when a company blocked external generative AI and the pressure to “bring some solution to the company” came from both above and below. That pressure is useful if channeled.
• Offer sanctioned, internal tools that mirror what people were using informally, but with controls and auditing. The goal is not to kill innovation, it’s to legitimize it, so people stop whispering about hacks and start sharing use cases openly.
• Document guidelines that answer the practical question every employee asks first, what data can I put into this tool. Clear rules remove the fear of punishment and reduce the chance of accidental exposure.
• Train, then train again. Ben emphasized hands on experience more than theoretical slides, show people “what generative AI can do” and then let them imagine how it maps to their tasks.
• Showcase internal examples. Nothing beats seeing someone in your department solving a real problem with AI. Ben’s approach was to create internal showcases and small videos, “so they really get to see this person who’s also in my business unit or who has a similar job that I have, he or she use it for this task.”
• Fast track adoption by example, not memo. Have senior leaders demo tools in public. If the boss of the boss is using it, people suddenly feel “it’s okay if I use it too.”

When not to use generative AI

Early in the hype cycle, people shoved everything at large models. Ben was blunt, some tasks were simply better solved using existing, mature tools. If you can perform a deterministic calculation in Excel with well-tested functions, don’t reinvent that with a language model. “Don’t use it to solve problems which you can already solve with existing technologies,” Ben advised. Today the lines are blurrier because models can orchestrate tools, call Python functions, and integrate with existing stacks, but the principle remains, choose the right tool for the job.

The cultural piece, curiosity beats FOMO

One of the most interesting behavioral problems isn’t risk aversion, it’s the opposite, FOMO. People who tried early versions and got garbage outputs often dismissed the whole field. Then the noise machine, analysts, and vendors shout about breakthroughs and people think it’s marketing noise. How do you get them curious again, not scared, not panicked, but curious enough to try the updated tools? Ben framed this as regaining trust, not convincing people to follow hype. “Maybe we need to invest a bit of time thinking, how do we spark curiosity in people?” That resonated. Curiosity is a gentler lever than fear of missing out.

A belief worth holding

If your company can combine top down signals with bottom up experimentation, create safe internal channels for use, and actually show employees useful wins that map to real tasks, trust will grow. It takes patience and iteration, because trust is not a policy, it’s a habit. Ben compared it to cars, how people regained trust in new car technology over decades. We don’t have decades to wait, but we do have playbooks and infra teams and, if used right, a fast path to responsible adoption.

So if you’re running the show, don’t try to solve this with one decree. Build some guardrails, fund a few small experiments, invite the skeptics to watch the demos, and most importantly, invest in curiosity. The future where AI helps people do more meaningful work is already within reach, and it starts with a company willing to replace whispered hacks with transparent, sanctioned, and useful practice.