Crisis of Trust is Why Customers Doubt Your Startup

On August 15, 2026, investor Gavin Baker went on X and told the CEO of a company about to become one of the most valuable in the world that he was doing it wrong. Baker’s argument, which he had also made on the All-In podcast, was that Dario Amodei’s constant warnings about AI risk had helped turn the public against the industry, and that Anthropic’s CEO owed his own sector a more positive pitch.

Amodei didn’t take the note. He agreed the public has a negative view of AI and called that a big problem. He just refused the diagnosis.

“I think it is fundamentally a crisis of trust,” Amodei wrote. “I think that ordinary people don’t trust companies, governments, or the tech industry and always suspect that we are cooking up some new way to screw them over.”

Then he said the part most founders skip. A crisis of trust is what happens when customers stop evaluating your specific claims and start assuming that anyone in your category is probably working an angle, which means better messaging cannot fix it and only delivered results can. That distinction is the whole game for any founder selling into a skeptical market, and it is worth a lot more than another brand refresh.

Last updated: August 2026

Quick answers

What is a crisis of trust?

A crisis of trust is when skepticism shifts from a specific company to an entire category, so customers discount claims by default rather than evaluating them on merit. It differs from a reputation problem because nothing you say about yourself resolves it. Only demonstrated, verifiable results move the number.

What did Dario Amodei say about the AI backlash?

Amodei said on August 15, 2026 that the AI backlash is “fundamentally a crisis of trust” rooted in decades of public distrust toward companies, governments, and tech, not in AI executives warning about risk. He added that the most accurate criticism of AI companies is that they haven’t yet delivered on their promises.

How do founders build trust with skeptical customers?

Founders earn trust by making claims narrow enough to verify, publishing evidence a customer can check without asking, and shipping consistently over a period long enough to establish a pattern. Buffer, Backblaze, and other transparency-first companies built credibility by releasing real numbers rather than describing their values.

What is a crisis of trust?

A crisis of trust is category-level skepticism, where customers stop assessing individual companies and start assuming bad faith from the whole industry. That’s the structural difference between a trust crisis and a bad quarter of reviews. A reputation problem is about you. A trust crisis is about the water you’re swimming in.

The distinction matters because the fixes are opposite. Reputation problems respond to communication: better positioning, clearer messaging, a founder who shows up and explains. Trust crises are immune to communication, because the customer’s prior is that communication from your category is manipulation.

Amodei made this concrete with a hypothetical. He said an ad claiming AI will cure cancer would likely be read as deceptive, so the promise is “more a cliche than it is inspiring.” His conclusion is the one line worth taping to a wall: “The thing that will work is actually curing cancer.”

You don’t have to be building frontier models like Anthropic for this to apply. Any founder in crypto, supplements, direct-to-consumer finance, weight loss, home services, or AI tooling is selling into a category where a meaningful share of buyers has already been burned, either personally or by proxy. Your claims arrive pre-discounted.

And the pool of low-trust categories keeps growing. Edelman found trust falling across nearly every institution it measures, including education, health care, and financial services, with social media the lone business category to tick up year over year while still holding the lowest score overall.

What Dario Amodei actually said about AI and trust

Amodei’s core claim was that AI skepticism predates AI. He rejected the idea that his messaging has been “disproportionately negative,” noting his writing has been “about equally balanced between risks and benefits” and pointing to his 2024 essay Machines of Loving Grace as the optimistic half of the ledger.

Where he conceded ground is the part founders should read twice.

“I think by far the most accurate criticism of AI companies including Anthropic is that we haven’t yet delivered on our big promises to benefit the world,” Amodei said, according to TechCrunch. “That is totally on us, and I think it’s the criticism you should be making, instead of all this stuff about messaging and marketing.”

A CEO five weeks from a reported October IPO publicly told his industry’s loudest cheerleader that the marketing critique was the wrong one and the delivery critique was the right one. That’s a rare thing to say out loud when you’re raising money. It’s also the correct read, and GJ has traced how Amodei got here in Anthropic’s IPO story and in our breakdown of how his stake was built.

Why don’t people trust AI companies?

People don’t trust AI companies because the category’s promises have consistently outrun its delivered results, and because trust in institutions generally has been falling for years. Edelman’s 2026 Trust Barometer, shared with Semafor, measured a three point global drop in trust in technology companies year over year, with US respondents landing roughly 30 points below China. Edelman’s tech-sector work has tracked US trust in AI companies falling from 50% to 35% across five years.

Now put that against Anthropic’s numbers. The company posted more than $11.5 billion in Q2 2026 revenue against $787 million in the same quarter of 2025, a jump of more than 14-fold, and told investors its annualized run rate hit $65 billion in July, per CNBC. That’s a long way from the $18.4 billion valuation it carried a few years ago.

Revenue is climbing while trust falls. Those two lines can diverge for a long time, because paying for a tool and believing in the company that makes it are different decisions. Enterprises buy Claude because it does the work. That purchase says nothing about whether a voter in Ohio believes Anthropic has her interests in mind when it builds a data center nearby.

Founders make a specific mistake here: they read revenue growth as proof that the trust question is solved. It isn’t. Revenue measures whether your product works today. Trust measures whether people will give you permission to keep operating tomorrow, which is what shows up as regulation, local opposition, churn during a scandal, and the price of every future launch.

Why your marketing cannot fix a trust problem

Marketing cannot fix a trust problem because marketing is the exact channel the skeptical customer has learned to discount. Every additional dollar you spend telling people you’re trustworthy confirms their model of you as someone who spends money telling people things.

This is why the “we need to tell our story better” reflex fails in skeptical categories. The story isn’t the bottleneck. The evidence is.

Baker’s advice to Amodei was essentially a comms fix: be a more positive advocate. Amodei’s counter was that positivity from an interested party is worth roughly zero when the audience already assumes interested parties lie. He’s describing the same wall a Series A founder hits when a prospect nods through the demo and then asks who else is using it.

There’s a related trap worth naming. Founders in low-trust categories often respond by getting louder about values: mission pages, manifestos, a founder letter about why this time is different. That content costs almost nothing to produce, which is precisely why it carries almost no signal. Anyone can write it. We covered the market’s growing allergy to cheap, unfalsifiable content in our piece on entity authority.

How do founders build trust with skeptical customers?

Founders build trust with skeptical customers by replacing assertions with artifacts a customer can verify independently, then repeating that over enough cycles to establish a pattern. The mechanism is cost. A trust signal only works if it would have been expensive or impossible to fake.

Buffer is the cleanest example of this. In December 2013 the company published every employee’s salary and the formula behind it. Applications for open roles went from 1,263 in the prior month to 2,866 in the month after, according to Buffer’s own write-up. Buffer didn’t claim to be fair. It published a number anyone could check and let candidates draw the conclusion.

Backblaze did the same thing with a harder number. Since April 2013 the company has published quarterly Drive Stats covering the annualized failure rates of every hard drive in its data centers, more than 330,000 drives as of mid-2025, released as an open dataset. A cloud storage company voluntarily publishing how often its drives die is a claim that would be very stupid to make if the drives were dying unusually often. That’s what makes it credible.

This also shows up in exit multiples. Buyers pay more for a business whose numbers they can verify without a forensic diligence process, which is a large part of the gap we documented between a 5.96x and a 2.24x micro SaaS sale.

Here’s how the common trust mechanisms compare on the only axis that matters, which is how hard they are to fake.

The pattern holds across every row. The mechanisms that work are the ones that would hurt if you were lying. If your trust play costs you nothing to run, your skeptical customer already knows that, and prices it accordingly. Founders raising against this backdrop should note that investors run the same test, which is part of why certain claims read as pitch meeting red flags.

What over-promising costs you in a skeptical category

Over-promising in a skeptical category costs more than it does in a trusting one, because each unmet claim is scored as confirmation rather than as a miss. In a high-trust market, a missed deadline is a mistake. In a low-trust market, it’s evidence.

Amodei’s cancer example is precise about this. The promise doesn’t fail because it’s too ambitious. It fails because the audience has heard versions of it before from people who didn’t deliver, so the ambition itself now reads as a tell.

The practical rule: shrink the claim until you can prove it, then let the proof do the expanding. A seed-stage founder saying the product cuts a workflow from 40 minutes to 6 for a named design agency is making a smaller claim than “we’re reinventing how teams work,” and a much stronger one. Backblaze never promised the most reliable storage in the industry. It published failure rates quarterly and let buyers reach that conclusion themselves.

Anthropic is now paying the bill on the other approach. Amodei conceded the industry’s big promises haven’t landed yet, and that admission is only necessary because the promises were made first.

This gets harder as buying moves to machines. In agentic commerce, an AI shopping agent comparing vendors has no way to be charmed by your mission page. It reads specifications, prices, and published performance data, which means unverifiable claims don’t just fail to persuade. They fail to parse.

Watch what this does to your roadmap. If every public claim has to be verifiable, you stop shipping announcements and start shipping receipts. That’s slower. It also compounds, because each verified claim raises the credibility of the next one, while each unverified claim taxes it.

How to handle public skepticism without getting defensive

Handle public skepticism by conceding the strongest version of the criticism and then narrowing your response to what you’ll actually do about it. Amodei’s exchange with Baker is a usable template, and it took him three moves.

He disputed the specific factual claim, which was that his messaging caused the backlash. He conceded the broader point, agreeing the public view is negative and that this is a real problem. Then he named the criticism he thought was correct and accepted it without hedging: the promises haven’t been delivered, and that’s on us.

Most founders do the opposite. They dispute everything, which signals that no criticism could ever land, or they apologize for everything, which signals they don’t know which part was true. Both read as evasion to a skeptical audience.

The move that fails hardest is the one Baker recommended and Amodei declined. Being a louder advocate for your own industry, at the exact moment the public is questioning that industry, is the least credible speech act available to you. Sam Altman and OpenAI have spent a similar stretch discovering that trust questions don’t yield to enthusiasm.

Trust is slow, boring, and cumulative. It gets built in the gap between what you said you’d do and what shipped, measured over and over by people who are not inclined to give you the benefit of the doubt. There’s no campaign for it. The thing that works is actually curing cancer, or whatever the much smaller version of that is in your business, and then letting someone else be the one to say so.