The people running the biggest AI companies cannot agree on how careful the industry should be. Anthropic's Dario Amodei wants a slower pace toward more powerful AI. Nvidia's Jensen Huang and Meta's Mark Zuckerberg say the market will keep things safe without anyone hitting the brakes. That argument, playing out on stages and in blog posts, is about frontier AI: the most advanced systems being built in research labs, not the tools running inside a typical company.
But offices do not experience that distinction. Employees, managers, and boards read the same headlines about AI systems becoming hard to control, then look at the chatbot summarizing meeting notes on their own laptop and wonder if it deserves the same worry. For a CIO, that confusion is now a real problem, separate from whether the AI itself works.
There is data showing the trust gap is not just perception. A survey of 202 senior AI decision makers at large US companies found that 98 percent have formal AI governance policies. Yet close to half admitted they had bypassed that same process when a deployment felt urgent. Other findings from the same research: more than a third of these companies had already suffered an AI incident serious enough to cause financial or reputational damage, and roughly half of those using highly autonomous AI systems had not updated their rules to cover the new risks those systems create.
A policy nobody follows under pressure is worse than no policy, because it invites the exact headline a company is trying to avoid. That is the real lesson for any manager watching their own company roll out AI tools: ask not whether a governance policy exists, but whether anyone can point to a time it was actually used to say no.
Saying no, in fact, may be the most useful thing a company can show right now. Advisors working with CIOs argue that a list of AI projects that got rejected, paused, or pulled back after launch does more to build trust than any list of successful rollouts. It proves the company is not deploying AI just because it is available, which matters directly to workers who are nervous about their own roles. Recent research on job fears puts the share of workers worried about being replaced by AI at close to half.
None of this settles the bigger argument between AI company founders about how fast the technology should move. A single company cannot fix that debate, and pretending otherwise will not build any trust. What a company can control is smaller and more useful: showing its own decisions, including the ones where it chose not to use AI at all, and being honest when something did not go as planned.
That is a lower bar than promising AI will never fail. It is also a far more achievable one, and it is the only kind of trust that survives the next uncomfortable headline.