AI Doesn’t Just Have a Trust Deficit. It May Have a Trust-Resistance Problem.

Sentiment and trust

Sentiment is not trust.

Anthropic has already done meaningful and increasingly sophisticated work to understand how people think about AI. Its first Anthropic Public Record surveyed nearly 52,000 Americans. Among its findings, only 15% said they trusted AI companies to make decisions about how AI is developed and used. Job loss was the most common fear, and respondents showed broad support for accountability and government involvement in areas including privacy, child safety and liability for harm.

Anthropic has also gone beyond large-scale survey research. Its broader public-input efforts have included interviews, in-person focus groups and conversations with people examining AI from different professional, cultural and institutional perspectives. That work matters. And it points toward an even deeper challenge.

Knowing what people think or feel does not necessarily tell us everything we need to know about trust. An institution may understand optimism, anxiety, adoption, skepticism and support for regulation while still lacking a complete picture of the beliefs underneath those reactions. Trust is connected to what people believe is true about an institution, how those beliefs were formed and what those beliefs are doing to behavior.

That distinction matters because sentiment can tell us something important about the temperature. Trust Intelligence helps us understand the conditions producing it.

Beliefs about institutions rarely form in a vacuum. They develop through what people experience, observe, remember and come to accept as proof.

Proof

AI may have a proof problem.

AI companies make enormous claims about the future. The technology will increase productivity. It will improve healthcare and accelerate science. It will create new economic opportunities. It will make people’s lives easier. Some of those outcomes may very well happen.

But from a trust perspective, the strategic challenge is not simply whether an institution can explain those possibilities convincingly. It is whether the world people experience begins to provide credible evidence for them. What would actually have to become true for those promises to feel proven?

That is a very different problem. If workers believe the economic upside of AI will accrue primarily to companies while they absorb the disruption, another report about productivity may not fundamentally change what they believe. If communities believe they are carrying the costs of technological expansion without participating meaningfully in its benefits, explaining those benefits more clearly may not be enough. If people are concerned about accountability when AI causes harm, better messaging about responsible development may not resolve the underlying concern.

The proof may have to become visible. Anthropic’s own public research offers an interesting signal. When Americans were asked what would help ensure AI benefits humanity, two of the highest-ranked responses were legal liability for harms and prioritizing safety over growth.

That does not tell us everything about trust. But it suggests something important. People may not simply be asking AI companies to tell a better story. They may be looking for proof.

And in a trust-resistant environment, institutional self-assertion becomes less persuasive. If the company building the technology says, this will be good for you, a skeptical stakeholder may reasonably think:

Of course you would say that.

The problem then becomes larger than the claim itself. It becomes a question of whether credible proof exists at all — and whether the institution has earned enough credibility for its own claims to carry weight.

How a claim becomes part of the trust environment

Institutional footprint

People do not experience an AI company as a chatbot.

There is another reason the trust question is bigger than product sentiment. People experience institutions through their total footprint. Artificial intelligence is already touching employment, education, creative work, science, healthcare, public policy, economic opportunity and daily life. Anthropic itself has begun publicly asking difficult questions about AI’s effects on jobs, families, creativity, safety, science and society.

That means the relevant trust question cannot simply be:

Do people trust Claude?

People are interpreting far more than what the technology can do. They are interpreting the institution behind it: its decisions, conduct, promises, visible priorities, and whether what it does provides credible evidence for what it says. That entire institutional footprint becomes part of the proof people use to decide what they believe.

Those beliefs can influence willingness to adopt, support, advocate for, regulate, resist or forgive the institution. This is why institutional trust cannot be reduced to brand perception. The entire footprint produces evidence. And people are constantly interpreting it.

The institutional footprint extends through: Product. Work. Education. Science. Policy. Infrastructure. Healthcare. Economic opportunity. Creative life. Everyday behavior.

Institutional response

Trust Intelligence has to change more than the dashboard.

There is also a danger in simply telling institutions that they need deeper research. More intelligence is not automatically better. An organization can ask sophisticated questions, conduct excellent research and learn extraordinary amounts about the people affected by its decisions. Then it can use everything it learned to become more effective at securing approval while changing nothing meaningful about the relationship.

That is not the purpose of Trust Intelligence. The goal is not to learn enough about stakeholders to make them feel differently about an institution. It is to understand the relationship well enough to determine what both stakeholder belief and institutional behavior are telling us about it.

That distinction is important. Because sometimes mistrust is not an information failure.

In those situations, the answer cannot simply be to change the interpretation. Something creating the interpretation may need to change. That may mean a change in decisions, commitments, incentives, accountability, the distribution of benefit or the way stakeholder intelligence reaches leadership.

This is where Trust Intelligence becomes consequential. It has to be capable of reaching the people who make decisions about how an institution operates. Otherwise, organizations risk creating increasingly sophisticated systems for listening to people without allowing what they hear to materially influence anything. And if the only thing that changes is public sentiment, the relationship itself has not necessarily changed.

Trust Intelligence is not always going to tell leadership how to communicate differently. Sometimes it should tell leadership to operate differently.

Organizational consequence

Anthropic is asking an important question.

Anthropic deserves credit for publicly engaging this problem. Beyond its Public Record research, the company has created a broader “Hard Questions” initiative, convened conversations around the social consequences of AI and recently appointed Mariano-Florentino “Tino” Cuéllar as its first Chief Global Affairs Officer, with a role that includes engagement with governments, civil society and community groups navigating the risks and opportunities created by advanced AI. That is encouraging.

But this conversation should extend beyond Anthropic. Every institution developing technology with the potential to reshape how people work, learn, create, participate and live will eventually confront some version of the same challenge.

The issue is not simply how to get people to trust an institution. It is whether leadership truly understands the trust environment surrounding it — and whether the institution is willing to examine its own contribution to that environment. That distinction becomes especially important when trust resistance is present.

Because the answer may not be a more persuasive explanation of what the institution intends to do. The answer may depend on what the institution is willing to make true. And what would actually need to become true, and visibly true, for trust to become warranted?

That is the harder question. Because trust is not awarded simply because an institution explains itself well. Trust follows what people believe has been demonstrated.

At Icon Element, we call the discipline of understanding those beliefs, the proof beneath them, their behavioral consequences and their implications for leadership Trust Intelligence. For AI companies, that intelligence may become increasingly important.

Not because the industry needs a more sophisticated way to convince people to trust it. But because institutions building some of the most consequential technology of our time need to understand whether the relationships surrounding that technology actually warrant the trust they are asking people to give.