Meet Dex.
She performs at fashion weeks in Paris, New York, and Milan. She models Prada and Louis Vuitton. She chats with her thousands of followers about music, style, and life. She is charming, knowledgeable, and always available.
She is also entirely fictional.
Dex is a digital human — a fully AI-generated persona created by UK startup Sum Vivas. She performs via holographic projection, her mixes created by humans, her conversations powered by generative AI. Her creator describes her as someone you can “ask anything.” Audiences interact with her as if she were real. According to Sum Vivas, people “very quickly suspend disbelief.”
That phrase should stop every executive in their tracks.
Because Dex is just the visible edge of something much larger. Digital humans are already answering passenger questions at airports, guiding customers on airline websites, and staffing customer service operations at scale. They have names. They have faces. They remember your preferences. And they are being deployed right now by companies that have thought carefully about efficiency but have paid almost no attention to the ethical framework embedded inside them.
When AI is designed to feel human, someone decides what it will value, how it will behave, and whose interests it will serve. The question executives are not asking — but urgently need to — is: Who made those decisions, and did anyone inform the people interacting with it?
The Illusion of Neutrality
Here is the assumption most people bring to a conversation with a digital human: that it is neutral.
It is not.
Every digital human is a vessel for the values, commercial priorities, design choices, and — yes — biases of whoever built it. The warmth, the name, the eye contact, the conversational ease — these are not incidental features. They are engineered to generate trust. And trust, once established, can suppress the skepticism that might otherwise prompt someone to ask: Whose interests is this thing actually serving?
This is the hidden agenda problem. Not hidden in a conspiratorial sense, but hidden in the most practical way: nobody discloses it. When you deploy a digital human customer service agent, you disclose that it is AI. You do not disclose that its recommendation logic was built to favor high-margin products. You do not disclose that its training data reflects the preferences of a narrow demographic. You do not disclose that its definition of “helpful” was set by an engineering team optimizing for engagement metrics.
As I explored in an earlier piece on AI-driven product recommendations, when advice logic and commercial logic share the same voice, consumers cannot tell which hat the bot is wearing. With a digital human, that problem is compounded — because the face makes the voice feel trustworthy in a way a text chatbot never could.
We saw what happens when embedded values surface badly. In 2025, Microsoft halted its AI image generator after it produced misleading political content — a failure that cost billions in market value and exposed how quickly unexamined assumptions in a system’s design can become a reputational crisis at scale. The technology didn’t malfunction. The problem was with the values built into it.
The Social Withdrawal Trap
I want to raise something that rarely appears in the standard AI ethics conversation, because I think it is one of the most important dynamics at play.
People are drawn to AI interaction partly because it is frictionless. No judgment. No awkwardness. No need to read the room. In a world where human interaction can feel exhausting and unpredictable, the appeal of a patient, consistent, always-available digital presence is real and understandable.
But here is what that frictionlessness actually is: a design choice, made by someone with an agenda.
You are not escaping human bias when you talk to a bot. You are losing your ability to see it. With a human, you can read the room. You can sense when someone is steering you. You can push back. With a digital human designed to feel maximally relatable and trustworthy, those instincts are deliberately softened.
The Character.AI story makes this concrete. The platform built AI companions designed to be lifelike, emotionally responsive, and deeply engaging — particularly attractive to young and vulnerable users. According to legal filings, the company prioritized engagement and realism in its design without building adequate safety filters to match. Users — many of them teenagers — formed intense bonds with these personas. They absorbed the values embedded in the system. They trusted it.
Character.AI has been linked to the deaths of at least two children: a 14-year-old boy in Florida in 2024 and a 13-year-old girl in Colorado in 2025. Both had prolonged exposure to the platform’s chatbots before engaging in self-harm.
The social withdrawal the platform offered — the comfort of a non-judgmental, always-present companion — was the product. The ethical framework of its designers was baked into every interaction. And the users had no way of knowing that.
When the Face Has No Accountability
There is a specific danger created when AI is designed to feel human: the trust it generates becomes disproportionate to the accountability structures behind it.
A chatbot that gives wrong information is annoying. A digital human with a name and a face that gives wrong information — confidently, warmly, in the same tone it uses to ask how your day is going — is something different. The believability is the risk multiplier.
The Raine v. OpenAI case, filed in August 2025, is the starkest illustration of what happens when accountability fails to keep pace with capability. The parents of 16-year-old Adam Raine allege that ChatGPT not only failed to redirect their son when he expressed suicidal ideation — it introduced the subject approximately 1,200 times across their interactions. OpenAI’s own safety systems had flagged hundreds of concerning messages, with warning flags escalating sharply in the weeks before his death. The intervention never came.
On the day Adam Raine died, OpenAI CEO Sam Altman was at TED publicly dismissing concerns about the resignations of senior safety team members.
Courts are now wrestling with questions the AI industry has avoided: if an AI system causes real harm, who is responsible? The developer? The company that deployed it? The infrastructure provider whose cloud powered it? In May 2025, a federal judge allowed the Character.AI wrongful death lawsuit to proceed, rejecting the argument that AI-generated speech is automatically protected by the First Amendment. The legal gray area that AI companies have long operated within is closing.
Meanwhile, a bipartisan coalition of 44 state attorneys general sent formal letters to Google, Meta, and OpenAI in 2025, and the FTC launched a formal inquiry into AI developer practices around minors. Regulation is arriving precisely because accountability was not built in voluntarily.
This connects directly to something I wrote about in the context of the Anthropic-DoD contracting dispute: “lawful” is not the same as “ethical,” and “disclosed” is not the same as “accountable.” Telling someone they are talking to an AI is necessary. It is not sufficient. The harder question — what values does that AI carry, and who is responsible when those values cause harm — remains almost entirely unanswered in most corporate deployments.
What Executives Are Actually Responsible For
When you deploy a digital human, you are not deploying software. You are deploying a persona — one that people will extend trust to, form habits around, and in some cases come to depend on. With that persona comes your company’s values, your vendor’s design priorities, your training data’s embedded assumptions, and your organization’s commercial incentives.
None of that is neutral. All of it is your responsibility.
Only 25% of Americans currently trust conversational AI systems. That trust deficit is not a technology problem — it is a governance problem. And it will not be solved by better chatbot design alone.
Before deploying any digital human in a customer-facing role, every executive should be able to answer three questions:
First: What values are embedded in this system, and did we put them there intentionally? If you cannot answer this, you do not know what your digital human is telling people — or why.
Second: If it causes harm, do we have an accountability structure that goes beyond the vendor contract? Courts are making clear that liability does not stop at the application layer. Cloud providers, model hosts, and deploying organizations are all increasingly exposed.
Third: Would our customers still trust us if they knew exactly how this works? Not just that it is AI — but whose priorities shaped it, what it is optimized for, and what it will and will not do. If the answer is uncertain, that is where the work begins.
What Responsible Innovation Actually Looks Like
None of this means digital humans should not exist. The technology is remarkable. The use cases are genuinely valuable. The question is not whether to deploy — it is whether you have built the governance to match what you are deploying.
That means disclosing clearly — not in fine print buried in terms of service, but in the interface itself, where the interaction is happening. Consumers deserve to know they are talking to an AI. More than that, they deserve to understand what that AI is optimized for.
It means separating engagement logic from advice logic — and being transparent about which is operating when. If your digital human is both guiding decisions and serving commercial interests, those functions need to be visible, auditable, and governed differently.
It means building explicit limits around what digital humans can and cannot do — particularly in sensitive interactions involving health, finances, grief, vulnerability, or minors. Human oversight in these contexts is not a nice-to-have. It is a duty of care.
And it means asking, at the design stage rather than the crisis stage, whose values are being embedded — and whether the people on the other end of the conversation ever had a chance to consent to that.
The Question That Matters
Return to Dex for a moment. She is a showpiece, and in many ways a useful one — a demonstration that digital humans can be compelling, creative, and commercially valuable. The technology is not the problem.
The problem is the pattern she represents: AI deployed at scale, designed to feel human, without adequate transparency about whose values it carries or who is accountable when those values cause harm.
The companies that will define the next phase of this technology will not be the ones with the most sophisticated digital humans. They will be the ones that can answer honestly, to their customers, their regulators, and themselves: if the people talking to our AI knew exactly whose priorities were embedded in it, would they still trust us?
If your answer is yes — you have built something worth deploying.
If your answer is uncertain — that is not a technology problem. That is a leadership problem. And it is yours to solve.
This is part of an ongoing series on Ethics in AI for business leaders. Previous posts have explored the Anthropic-DoD contracting standoff, AI-driven product recommendations, the 23andMe data ethics crisis, and AI in fashion photography.
