The first sign that someone had stolen my voice came at 9:07 on a Wednesday morning, when a client emailed me, “Thanks again for yesterday’s call, Natalie. Your promise that the migration would be completed by Friday really put our board at ease.”
My name is Natalie Carter, I was thirty-six, and I worked as a senior implementation manager for a healthcare-software company in Boston. I had not spoken to that client in eleven days, and there was absolutely no chance I had promised a Friday migration because our engineering team had already documented that the earliest safe completion date was three weeks away.
I replied carefully.
“I wasn’t on a call yesterday. Could you tell me who attended?”
The client, Michael Reeves, answered within minutes.
“You were there with Eric. I spoke directly to you.”
Eric Lawson sat twelve desks from me.
He had worked beside me for two years.
I walked over and asked, “Did you have a call with Reeves Medical yesterday?”
Eric barely looked away from his screen.
“Yeah. Why?”
“He says I was there.”
Eric paused.
“Maybe he’s confused.”
Michael was not confused.
At my request, he sent me the recording from their videoconferencing platform, and I put on headphones at my desk expecting some obvious misunderstanding.
Then I heard myself say, “Michael, I personally guarantee we’ll have everything migrated by Friday.”
My hands went cold.
The cadence was mine.
The slight rasp I got when I spoke quickly was mine.
Even the way “personally” rose at the end sounded exactly like me.
But I had never said those words.
The recording showed Eric on camera while the other participant appeared only as a generic black square labeled Natalie Carter.
I replayed it four times.
Then I noticed something subtler.
The fake version of me never interrupted anyone, never coughed, never stumbled over a sentence, and answered every question after almost exactly the same pause.
It sounded human until you knew where to listen.
I immediately contacted my director, Hannah Brooks, and information security. Before they could respond, another client called.
“Natalie, are we still waiving the $18,000 integration fee like you said Monday?”
I nearly dropped my phone.
“What Monday call?”
Silence.
Then the client said, “The one with Eric.”
By noon, security had found six meetings on Eric’s calendar where my name appeared as an attendee even though my account showed no login activity.
Three had recordings.
In all three, my voice was there.
By 2:00 p.m., Eric’s access had been suspended.
At 2:17, he walked past my desk with security beside him and stared directly at me.
“You’re blowing this completely out of proportion.”
I stood.
“You used my voice.”
He stopped.
Then he said something that made the entire office go silent.
“Your voice was already company data.”
Information security isolated Eric’s laptop that afternoon, while legal instructed me not to contact any affected clients until they understood exactly what had happened. I sat in a conference room listening to three recordings of a synthetic version of myself making promises I would never have approved.
The worst one was not the waived fee.
It was a meeting with a hospital group in Connecticut where “I” assured them that a security feature would be available before their launch.
It was still in development.
“If they relied on that statement,” Hannah said quietly, “we have a serious problem.”
The forensic review revealed that Eric had collected more than eleven hours of my speech from recorded internal meetings, training sessions, webinars, and client presentations. He had then used commercially available voice-cloning software to create a synthetic version convincing enough to fool people who already knew me professionally.
But the investigation uncovered something even more disturbing.
He had not started with clients.
For almost three months, Eric had tested the cloned voice on internal calls.
He would join meetings early, tell participants that my camera was broken, and trigger short prerecorded or generated responses such as, “I agree with Eric’s recommendation,” or, “Yes, we can support that.”
Most people never questioned it because the statements were brief.
And because they trusted me.
When legal interviewed Eric, he initially claimed he had only been “automating repetitive meetings.”
Then security found a folder containing scripts.
One was titled:
NATALIE — CLIENT OBJECTIONS
Another:
NATALIE — APPROVAL PHRASES
I stared at the filenames until Hannah finally closed the laptop.
“He wasn’t just reproducing you,” she said.
“He was borrowing your authority.”
That was exactly it.
Eric was a talented salesperson, but clients frequently pushed back when he made technical commitments. I had seniority, a reputation for refusing unrealistic deadlines, and enough credibility that when I promised something, people believed it.
So he had turned that reputation into a tool.
Then legal found the reason the behavior had suddenly escalated.
Eric’s quarterly bonus depended on closing two large renewals before the end of the month.
The accounts where my cloned voice appeared were worth more than $3 million combined.
That evening, Hannah called me into her office.
“There’s another issue.”
I expected another recording.
Instead, she turned her monitor toward me.
An internal email showed that Eric had sent one of the voice samples to our sales director, Mark Dalton, six weeks earlier with the subject line:
“AI Natalie — pretty convincing, right?”
Mark had replied:
“Terrifyingly good. Just don’t do anything stupid with it.”
I stared at Hannah.
“He knew?”
“He knew the clone existed.”
“And he never told me?”
“No.”
For the first time, this stopped feeling like one coworker crossing a line.
Someone higher up had seen the weapon being built and treated it like a joke.
Eric was fired two days later, but the company quickly learned that removing him was the simplest part of the problem.
Legal contacted every client whose meetings had been affected and explained that statements attributed to me might have been generated without my knowledge or approval. Two pricing commitments were withdrawn and renegotiated, while the hospital project was delayed rather than launching around a security feature that did not exist yet.
The company absorbed part of the financial loss.
That was appropriate.
The clients had trusted a voice presented as mine through company channels, and simply telling them they should have detected a synthetic speaker would have shifted responsibility onto the wrong people.
Mark, the sales director who had seen the demonstration weeks earlier, was placed on administrative leave during the review.
His defense was painfully ordinary.
He said he thought Eric was experimenting.
He said he never believed Eric would use the clone with customers.
He said, “Just don’t do anything stupid with it,” had clearly been a warning.
I asked him one question during the investigation.
“If someone showed you a fake key to my house, would you wait until they used it before mentioning it to me?”
He did not answer.
Mark ultimately kept his job, but lost management responsibility and received a formal disciplinary action for failing to escalate the issue. The company also introduced rules requiring consent before employee voices or likenesses could be used in generative systems, mandatory disclosure of synthetic audio in external communication, and technical verification for recorded client approvals.
Eric threatened to sue over his termination.
He never did.
Instead, through his attorney, he eventually admitted using the synthetic voice but argued that because the source recordings belonged to the company, he believed experimenting with them was permitted.
That argument stayed with me longer than anything else.
The recordings belonged to the company.
My identity did not.
For several months, I hated hearing myself in legitimate recordings. I would replay sentences and wonder whether other people now listened for glitches, unnatural pauses, or tiny signs that perhaps “Natalie” was not really Natalie.
The damage was strangely personal even though nothing physical had been stolen.
A password can be changed.
A credit card can be canceled.
I could not replace my voice.
Six months later, Michael Reeves—the first client who warned me without realizing it—visited our office.
Before leaving, he said, “I still feel stupid for believing it.”
“You shouldn’t.”
“It sounded exactly like you.”
“That was the point.”
He nodded, then smiled slightly.
“So how do I know this conversation is really happening?”
For the first time in months, I laughed.
“You’re standing three feet away from me.”
The joke helped, but the lesson remained.
Technology had not created Eric’s dishonesty.
It had simply allowed dishonesty to wear my identity.
And the most frightening part was not that artificial intelligence could imitate my voice.
It was that everyone trusted the imitation because I had spent years making the real one worth trusting.
That reputation had taken me a decade to build.
Eric copied the sound of it in an afternoon.
What he could never copy was the responsibility that was supposed to come with speaking in my name.
