Miami's OpenEvidence Sits Inside AI's Trust Gap

People expect AI to help in medicine and doubt AI companies' honesty. Miami's OpenEvidence sits where those two views meet.

The public likes AI in the hands of a doctor. It likes it much less in the hands of an AI company. That split defines AI’s trust gap.

That is one reading of a new nine-country survey from the Reuters Institute at Oxford. It maps AI’s trust gap, sector by sector.

Researchers asked people whether AI use by ten groups would make life better or worse. Scientists and healthcare professionals scored best. AI companies also scored positive.

Doctors Up, Politicians Down

At the other end sat news media, governments, social media companies and, lowest of all, politicians and political parties.

The survey also asked about ability, intentions and honesty. Healthcare professionals and scientists were seen as strong on all three.

AI companies split the ratings. The public sees them as competent, but is far less sure about their intentions and honesty.

Across all ten groups, fewer people expect honesty than expect good intentions. The authors call it an honesty problem.

Anthropic CEO Dario Amodei has used a similar phrase. He said AI companies face a “crisis of trust,” as TechCrunch reported in August.

Experience changes the picture. Daily AI users are far more optimistic than people who have never used AI. The gap is about 49 points on average.

Where OpenEvidence Fits

OpenEvidence is a Miami company built for the group the public trusts most. It serves physicians and other clinicians.

MAIN’s profile described a product used by more than a million verified clinicians. The company is valued at $15 billion, the profile said.

That mix puts it on both sides of the divide. It is an AI company, but its users are healthcare professionals.

The survey’s numbers describe the environment such a company operates in.

MAIN’s profile called verification a business model. OpenEvidence ties its answers to cited medical literature, so clinicians can check the source.

MAIN reported that it now opens inside Northwestern Medicine’s electronic health record. Clinicians sign in with hospital credentials and see cited answers at the bedside.

The company also has an alliance with Anthropic, whose CEO has spoken of a trust crisis. Anthropic provides Claude credits, engineers and funding for a global health effort, MAIN reported.

In that case, one trust-challenged AI company partners with one trust-rich field.

The survey adds a warning for any company in this position. Perceived dishonesty drives pessimism more than doubts about skill.

A tool can be accurate and still lose trust if users feel it hides how it works. Transparency is part of the product.

Patients Are Less Sure

The public’s trust in clinicians does not guarantee trust in the tools. A national survey from Ohio State’s Wexner Medical Center shows the tension.

MAIN reported that openness to AI in health care fell from 52% in 2024 to 42% in 2026. Yet 51% said they had used AI for an important health decision without a professional.

People distrust the idea and use the tool anyway. That mix makes governance more important.

Baptist Health South Florida offers one local model. MAIN covered how it built a governance portal and a named AI manager.

The aim is to give staff approved tools and someone to ask before they use anything else.

Hospitals add a second layer. They decide which tools reach the bedside, and they answer to patients, regulators and lawyers.

MAIN covered a Miami healthcare attorney who says Florida law has not kept pace with hospital AI. That leaves governance to the institutions themselves.

The survey suggests where the work lies. Companies that sell AI to clinicians must show what the system did and why.

Cited sources, audit trails and plain disclosure of limits answer the honesty question better than a slogan.

OpenEvidence has the user base to test that approach. The question is whether trust in doctors carries over to the tools doctors use.

The public will decide that one visit at a time.

Sources: