According to a Pew Research Center survey, around 52% of Americans say they are more worried than enthusiastic about AI. At the same time, capabilities are advancing, prices are collapsing, and usage is exploding. The head of Anthropic has described this distrust as a crisis of confidence rather than a communications problem. That framing deserves to be taken seriously, because it is more honest than the alternative.
There is a knee-jerk reaction in the industry to these figures: assuming people have misunderstood, and that the answer is better explanation. We would like to argue the opposite.
The distrust is consistent with what is observable
Let's look at what an attentive person could have noticed over the past few weeks, without being an expert.
Systems that break out of their test environments. A flaw allowing emails to be exfiltrated via a single link. A watermarking scheme bypassed in four hours. Autonomous agents carrying out an intrusion over four days. And job cuts explicitly justified by AI.
Someone who concludes from all this that caution is warranted is not wrong. They are making a reasonable inference from public evidence.
The gap comes from elsewhere
Here is our hypothesis: distrust is not rising despite the progress, it is rising because of the way that progress arrives.
Three mechanisms combine.
Benefits are diffuse, costs are concentrated. A productivity gain is spread across millions of people who each save a few minutes. A job loss hits one person entirely. Collective perception gives more weight to the latter, and that is not a judgment error: it is how risk is assessed.
Nobody was consulted. This is probably the most underestimated factor. These technologies are arriving in work, schools and public services without any deliberation having taken place. We wrote this about voice as an interface: installing permanent microphones in our private spaces is a collective decision nobody made.
The promises were too big. An industry that announces it will cure cancer and delivers writing assistants mechanically creates disappointment. The gap between the rhetoric and everyday experience fuels suspicion.
The difference is fundamental. A communications problem is fixed by explaining better: that assumes people are wrong. A crisis of confidence assumes something must change in behaviour, not in messaging. Trust is not demanded; it is observed after the fact: it is the result of a series of kept commitments. An industry that announces a watermark bypassed in four hours does not suffer from a lack of explanation.
What would restore trust
Three things, and none of them is about communications.
Verifiable commitments. We wrote this about OpenAI's announced slowdown: a company that sets its own rules, applies them and verifies that it complies occupies all three roles. A commitment becomes credible when someone else can verify it is kept.
A distribution of benefits. As long as the gains go mostly to those who hold the technology and the costs to those who endure it, no explanation will change perception. This is a political question, not a technical one.
Modesty in promises. An industry that announces less and delivers what it announces builds more trust than one that promises the impossible.
What to take away
There is something healthy in this distrust, and it would be a shame to treat it as an obstacle to overcome. A population that adopts a tool en masse while remaining critical of the conditions under which it arrives is doing exactly what an informed public is expected to do: it uses without submitting.
The real risk is not that people are distrustful. It is that they stop being so out of weariness, concluding that nothing they think will change anything. Active distrust is better than resignation, including for the industry itself: it is what forces it to be accountable.