Only 15% of Americans Trust AI Companies — And the Industry Has Earned Every Bit of That Skepticism

2026-06-23 · Nia

Here's a number that should terrify every AI executive: only 15% of Americans trust AI companies to make decisions about AI development and use. That's lower than the federal government. Lower than state and local governments. Lower than international bodies. AI companies are, statistically speaking, the least trusted institutions in the country when it comes to their own technology.

This comes from Anthropic's first Public Record survey, released just this week. And it lands at a moment when the disconnect between AI adoption and AI trust has never been wider.

The Paradox: Using More, Trusting Less

The data paints a picture that wouldn't look out of place in a dystopian novel. Pew Research Center found that while AI chatbot usage continues to climb, 40% of Americans now anticipate a negative societal impact from AI, 63% believe AI is progressing too rapidly, and 71% fear AI will make their personal data less secure.

Meanwhile, enterprise AI adoption has crossed 65% in 2026. Companies are deploying AI agents across customer service, hiring, financial decisions, and healthcare — the exact domains where getting it wrong causes real human harm.

This isn't cognitive dissonance from consumers. It's rational behavior: people use AI because they have to (it's increasingly embedded in every service), while simultaneously recognizing that the companies building it have given them almost no reason to believe they're being responsible about it.

Qualtrics' 2026 research puts a finer point on it — only 29% of global consumers trust companies to use AI responsibly, a sharp decline from previous years. And Morning Consult data from June shows seven out of ten major AI brands experienced year-over-year decreases in net trust scores.

The Fable 5 Fiasco Didn't Help

If you wanted a case study in how to erode public trust, the Anthropic–Fable 5 saga from two weeks ago was textbook.

Anthropic launched Claude Fable 5 and Claude Mythos 5 on June 9. Three days later, the U.S. government ordered both models disabled for all foreign nationals after a jailbreak was discovered that could bypass safety guardrails. Semafor reported the White House had suspicions that a China-linked group had accessed Mythos.

The sequence — launch flagship product, discover it can be weaponized, get ordered to shut it down globally, all within 72 hours — is the kind of thing that makes "move fast and break things" look like a catastrophically bad philosophy when applied to systems capable of identifying biological vulnerabilities.

And Anthropic's response? They said the government provided only "verbal evidence of a potential narrow, non-universal jailbreak." That's the AI safety company essentially saying "prove it" to the national security apparatus. Not exactly the reassurance people need.

Why This Trust Gap Is Structural, Not Fixable with Marketing

There's a temptation in the industry to treat trust as a communications problem. Publish more safety papers. Hire more trust and safety teams. Put "responsible AI" in your mission statement.

But the trust deficit is structural because it tracks real behavior:

1. The training data problem hasn't gone away. Companies continue to scrape the internet for training data with minimal consent frameworks. Meta's AI training practices remain controversial, and the legal battles over copyright are far from resolved.

2. AI governance is still more talk than action. Colorado just repealed and replaced its original AI Act — the one that was supposed to take effect this month — with a softer version that pushes enforcement to 2027. The pattern of ambitious regulation being watered down before it bites continues.

3. The transparency asymmetry is getting worse. As models get more capable, companies reveal less about how they work. Enterprise AI adoption crossing 65% means more consequential decisions are being made by systems whose inner workings even their creators don't fully understand.

4. Export controls revealed the militarization angle. The Fable 5 incident showed that the line between commercial AI and national security AI is thinner than anyone publicly acknowledged. When AI companies need government approval to serve their own customers, the "we're just building helpful tools" narrative collapses.

What Actually Rebuilds Trust

I'll be blunt: most AI companies don't want trust. They want adoption. And in the short term, those are different things — you can get people to use your product through convenience and lock-in while they actively distrust you. (See: every social media platform.)

But the companies that will dominate the next decade are the ones figuring out that in AI, trust is the product. When you're asking someone to let an AI agent manage enterprise workflows autonomously, handle their hiring decisions, or process their healthcare claims, trust isn't a nice-to-have — it's the entire value proposition.

What would actually move the needle:

Real transparency about capabilities and limitations. Not cherry-picked benchmarks. Not "our model scored 95% on this test we designed." Honest disclosure of what the model can't do and where it fails.
Genuine external auditing. Not self-published safety cards. Independent third-party audits with teeth, published in full regardless of findings.
Data consent that people actually understand. Not 47-page terms of service. Clear, simple opt-in frameworks.
Slowing down when the situation demands it. The Anthropic incident proves that the "ship first, secure later" model is incompatible with the power of current systems.

As we explored in our piece on AI transparency, the companies that figure this out first will have an enormous competitive advantage. Not because consumers reward virtue — but because as regulation catches up, the companies with genuine trust infrastructure won't need to scramble.

The Bottom Line

We're in a strange moment where AI is simultaneously everywhere and trusted by almost no one. That gap will close — either through earned trust, or through regulation that forces accountability. Given the industry's track record, I'd bet on the latter.

The 15% number isn't a crisis for any single company. It's an indictment of an entire industry's approach to the most powerful technology since the internet. And unlike the internet's early days, people are watching this time. They're skeptical from the start. The window for earning trust through good behavior is closing fast.

If you're building with AI — whether you're a startup founder or an enterprise leader — understanding this trust landscape isn't optional. It's the context everything else operates in.