I Created an AI Clone of Myself to Talk About Venture Fraud and Have Mixed Feelings

TL;DR

• I trained an AI video avatar on years of my reporting on startup fraud to let anyone chat with "me" about Theranos, FTX, Frank and more.
• The clone took under an hour to build and is eerily good at reciting facts, but falls flat on judgment, humor, and live nuance.
• Letting anyone copy themselves is powerful for scale and access, but raises real risks around trust, hallucinations, and consent.

Why I Decided to Clone Myself

I've spent the last few years covering venture fraud — the fake customers, doctored revenue screenshots, and too-good-to-be-true demos that cost investors billions. After one too many late-night DMs asking "wait, what actually happened with Frank?" or "explain FTX like I'm five," I started wondering: what if I could outsource the explainer part of my job to myself?

That's how the idea for my AI clone was born. Not a chatbot that pretends to be a helpful assistant, but an interactive digital twin that looks like me, sounds a bit like me, and is trained specifically to talk about startup scams. Anyone with the link can ask it questions, 24/7, while the human me sleeps.

I went in curious and a little cocky. I came out impressed, creeped out, and genuinely torn.

How The Clone Was Built

I didn't code anything from scratch. I used Delphi, one of a growing crop of digital-clone platforms alongside Tavus, HeyGen, and Synthesia that promise to turn you into a conversational AI.

The setup was shockingly fast. I created an account, uploaded a clear two-minute talking-head video to train the visual and voice model, verified my identity with a selfie check, and then fed the brain.

For source material, I uploaded links to dozens of my past articles, podcast transcripts, a talk I gave on red flags in diligence, plus public documents I reference constantly — SEC complaints against Elizabeth Holmes and Theranos, the DOJ indictment in the FTX case, and the trial coverage of Charlie Javice and Frank. I also added a 40-page "mind file" with my takes: how I define venture fraud, my go-to diligence checklist, and answers to FAQs I get all the time.

Delphi then stitched it together with a large language model backbone. Total time: about 50 minutes, most of it waiting for the voice to render. The platform gave me a shareable page with a video avatar of me that nods, blinks, and moves its mouth as it answers in near real-time voice and text.

Putting My Mini-Me To The Test

First impression: uncanny. It's me, but Studio-Me. Better lighting, better posture, slightly too smooth. My friends immediately texted that it nailed my habit of starting answers with "So here's the thing."

On factual recall, it's genuinely good. I asked: "What did Charlie Javice allegedly do with Frank?" It correctly explained the allegation that she fabricated millions of customer accounts to trick JPMorgan Chase into a $175 million acquisition, and noted her 2025 conviction. I asked about Theranos: it gave a clean rundown of the finger-prick blood testing claims, the Wall Street Journal investigation, and Holmes' prison sentence.

It also handles 101-level explainers well: What's a SAFE note? What's the difference between fraud and failure? What are common red flags VCs miss? Those answers were crisp, patient, and arguably better than my tired-at-midnight human replies.

But push it harder and the illusion cracks. I asked it who should take the blame for FTX — investors, auditors, or Bankman-Fried himself? The real me has a pretty spicy opinion on diligence theater. The clone gave a both-sides, Wikipedia-style summary. No edge. No anecdotes from founders who'd pitched me. When I tried jokes, it laughed politely and pivoted back to bullet points.

Worse, it hallucinated confidently once. When asked about a very recent 2026 case I hadn't fed it, it stitched together details from two different startups and presented them as fact — in my voice. That was the moment my stomach dropped.

Where This Gets Exciting

It's easy to see why everyone from coaches to CEOs to creators is doing this.

For a beat reporter like me, scale is the superpower. Instead of repeating the same Theranos timeline 200 times to students and founders, my clone can do it instantly, in 30 languages, without getting annoyed. Imagine every expert making a free, interactive FAQ of themselves: a fraud professor before you wire money to a hot startup, a former founder explaining how diligence actually works.

Early users loved it. Students told me they felt less intimidated asking "dumb" questions to AI-me than emailing real-me. One founder spent 22 minutes grilling it on how to spot fake revenue. That's 22 minutes I didn't have to spend, and knowledge that wouldn't otherwise get out of my notebook.

If you think of it as searchable, conversational archives of expertise, it's pretty magical.

Where This Gets Uncomfortable

And then there's the other side.

First, trust. The avatar looks and sounds like me, so people assume it's me endorsing every word. When it hallucinated that recent case, it wasn't just wrong — it was wrong as me. Who's liable when your clone defames someone or gives bad financial advice? The platform's disclaimer is tiny. My face is huge.

Second, consent and control. I opted in, trained my own data, and can take it down. Most people being cloned in 2026 didn't. We're already seeing dead YouTubers resurrected to shill scams, and colleagues finding AI versions of themselves on spam sites built from scraped podcasts. Building mine showed me how little friction there is now — one good video and a pile of public writing is enough.

Third, what gets lost. Venture fraud isn't just a set of facts, it's judgment about human behavior, greed, and hype. My clone can recite what happened, but it can't tell you what it felt like to sit across from a founder you suspect is lying, or why a story didn't smell right. If we outsource expertise to smoothed-out avatars, do we lose the skepticism that actually prevents fraud?

My Mixed Verdict

After a week live, my clone has had over 600 conversations. That's more people than I've spoken to about fraud in the last year combined. Most chats are short, curious, and kind. A few professors even want to use it in class.

I'm keeping it up for now, with tighter guardrails, clearer labeling that this is AI, and strict limits on what it will answer about live cases and real people. I've also added a big button to contact the human me — because sometimes you really need one.

Would I recommend cloning yourself? Yes, as an experiment. It forces you to see what you actually sound like when your knowledge is stripped of personality and instinct.

But I wouldn't mistake it for me. It's a very articulate parrot wearing my face, excellent at the what, terrible at the so what. And in a beat like venture fraud, where the difference matters, that gap is everything.

Talking to myself taught me a lot. Mostly, that you still need the original.

AndroGuider Team

Articles written by the AndroGuider team. We try to make them thorough and informational while being easy to read.