Human Sincerity In Digital World

There's a feeling most people know, but no one has a word for. You're listening to someone you have no reason to doubt — a colleague, a customer, a candidate — and they're telling you everything is fine. The words are right. Something beneath them isn't.

A sales rep reports the deal is "in good shape," and the confidence in their voice is exactly what makes you reach for the forecast with both hands. A customer on a renewal call says they're "happy with the product," and something in the way they say it is the first quiet signal that they're already evaluating competitors. A candidate claims they delivered phenomenal performance in their last role, but something about it feels exaggerated. 

In every one of those moments, you knew exactly who you were talking to. The gap was never about identity. It was the distance between what they were expressing and what they actually seemed to feel, and it's the question that none of the trust and safety tools you use today are built to answer. That gap has a name. We call it sincerity: the alignment between what someone expresses and what they appear to feel in the moment. It has nothing to do with knowing who someone is. It's the question left even after identity is settled, and it's the one most trust and safety tools today aren't built to answer.

The Gap Is Getting Harder to Spot, Not Easier

For most of human history, that gap was hard to close on your own. You read tone, body language, went with your gut, and sometimes you were wrong. But at least you knew who you were dealing with. A face and a voice were reliable proof that a real person was in front of you.

That assumption is now weaker. A finance employee at a company most of us recognize recently joined a video call with his CFO and several colleagues, and by the end, he had wired $25 million to an outside account. Everyone else on that call was a deepfake. Voice clones and fake video calls are getting cheaper and more convincing, and fraud like this is becoming more common.

Now there are two problems stacked. The old one, knowing whether a real person's expression matches what they feel, was never solved. The new one, which lets you know whether you're even talking to a real person, is spreading fast and can't be ignored.

Identity Solves Half the Problem

A lot of serious, valuable work is already underway to solve the new problem. Deepfake identification, confirming whether the person or media in front of you is real and unmanipulated, is advancing fast and needs to keep up. But even if it were fully solved, it would answer only one of the two questions on which online trust depends.

Confirming that someone is a real, unmanipulated human confirms their existence. It doesn't tell you whether what they're expressing in this moment matches what they actually feel. A verified person can still mislead you with how they say something, not just whether they're real.

Trevor Milton, the founder of electric truck maker Nikola, is a clear example of exactly this gap. He wasn't a deepfake. He was the real founder, on camera and under his own name, making confident public claims about his company's technology. Regulators later found that he knew those claims weren't accurate, including a claim that Nikola was producing hydrogen when it wasn't producing any. He was convicted of securities fraud. Identity verification would have passed him without hesitation, because he really was who he said he was. What failed wasn't his identity. It was the gap between what he said in public and what he already knew in private. Elizabeth Holmes and Theranos follow the same shape: a real, verified founder whose confident public claims about the technology didn't match what she privately knew it could actually do. Builder.ai followed the same shape from a different angle. The Microsoft-backed, $1.5 billion startup marketed an AI assistant, Natasha, that it said could build software autonomously, and it reported roughly $220 million in revenue for 2024. After the company collapsed into insolvency in 2025, an internal audit and a Bloomberg report found that the real figure was closer to $55 million, inflated through a round-tripping scheme with another firm, and that much of the work attributed to AI had been done by human engineers rather than by the autonomous system it sold. None of that was an identity problem. Builder.ai was a real company, verified and backed by some of the world’s most sophisticated investors. What was missing was the alignment between what it expressed and what was actually true.

Sincerity Is the Harder Half

As identity verification improves and spreads, the other question becomes harder, not smaller. Once you know you're talking to a real person, you're left with the gap we opened, the one that’s never about who someone is. It's about whether someone's way of communicating aligns with what's really happening with them. This is becoming more critical as fraud, like the $25 million call, shifts from “is this person real?” to “can I trust what a real person is telling me?”

Identity verification isn't the only driver. AI voice agents increasingly handle these conversations directly: customer service lines, answering services, first-pass sales calls. A synthetic voice alone isn't a fraud problem. Everyone on the call usually knows it's an agent. Making that voice sound convincingly human is its own discipline, including matching pace and tone, knowing when to escalate rather than repeat a scripted line. But sounding human isn't the same problem. Even a well-tuned agent can keep repeating the same reassurance while a caller's frustration builds. The gap between what it says and what's actually happening remains, no matter how polished the voice is. The more these conversations move to agents rather than people, the more this becomes the infrastructure every call needs, not a feature reserved for calls a person still handles.

Building Both Components

We think rebuilding trust in digital communication means building both components, not just the one that’s getting the most attention right now. Not a security feature bolted onto video calls after the fact. Something closer to infrastructure. Confirming a real human is present, then surfacing whether their expression matches what they appear to feel. One more signal before deciding how much weight to put on what they're saying.

We don't think either half gets solved with a single answer, and we're wary of solutions that ask everyone to give up their privacy to get there. But it's worth being clear about how big the full problem is, because it’s not just about stopping fraud. The same gap shows up in smaller, everyday ways too. A bot reply, a fake dating profile, or an incentivized review usually isn't an identity problem. The account is often real. What's missing is whether the person behind it is actually expressing what they mean. Imagine a version of the internet where replies aren't mostly bots, where a dating app match is reliably a real person, where a product review comes from someone who actually bought the thing. That's the upside of solving this, not just the downside of leaving it unsolved.

Knowing someone is real has never been the same question as knowing whether what they're expressing lines up with what they actually feel. Both still need an answer, not just one.