What Can a ChatGPT Voice Mode Hallucination Sound Like?
Forgot password
Enter the email address you used when you joined and we'll send you instructions to reset your password.
If you used Apple or Google to create your account, this process will create a password for your existing account.
This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.
Reset password instructions sent. If you have an account with us, you will receive an email within a few minutes.
Something went wrong. Try again or contact support if the problem persists.
Images by cottonbro studio on Pexels.

Woman uses ChatGPT voice mode during staycation. Then the AI allegedly stops talking and starts giggling with someone else: ‘Are we going insane?’

“It was like she was in an office.”

Natalie Elizabeth Ellis and her friend, Alexandria Wombwell-Povey, were recently playing around with ChatGPT during a staycation when they claimed the chatbot got creepy. The pair said they heard something unsettling during a voice mode response, and the experience left both of them shaken. Natalie took to Instagram to see if anyone else had experienced anything similar. The video has drawn over 335K views, but commenters offered theories rather than concrete answers.

Recommended Videos

In the video, Natalie told her viewers, “Okay, we’re really freaked out. We just had a crazy AI experience.” She explained that they were talking to the chatbot about religion when what she described as a glitch occurred. Natalie, who uses a female voice setting for the app, said she heard the AI stop talking entirely. Then, she claimed she heard a male voice say, “I could hear your conversation.” After that, the AI supposedly resumed its original female voice.

Alexandria said that it sounded a little different to her. “I heard… it was like she was in an office and she was responding to us, and then someone started to talk to her. So she started speaking, stopped, and then I could hear another voice. And then, she like giggled a bit.” Natalie immediately jumped in and agreed. Alexandria continued, “Yeah, she did. She giggled. That’s why it reminded me of being in an office. She like giggled. Then came back to us and carried on talking.”

They checked for other possible sources of sound

Natalie said, “We’re both so freaked out we feel nauseous. I swear to god. I heard a male voice saying, ‘I can hear your conversation,’ and so immediately we got freaked out. Came off the app, and I was like, ‘Oh, maybe the Nanit app must be on. Like the baby monitor. Check the Nanit… sound off.” 

Alexandria jumped in and claimed that when they asked ChatGPT about it, “she was like, ‘Oh it’s an audio glitch.’ But… I’m like, you were literally talking to someone else. Like i heard someone else’s voice, and she responded to it. She like giggled.” 

Natalie then ended the video by asking, “So okay, we need to know. Has anyone else had this experience? Are we going insane right now?” Of course, the account comes only from Natalie and Alexandria and cannot be independently verified. However, similar glitches have been reported over the past few years.

Two years ago, on the OpenAI subreddit, a user claimed that the AI laughed out of the blue and said something unintelligible. They wrote, “It genuinely gave me chills. I wish I could have recorded it on my iPhone to show what happened.” 

Bug reports on OpenAI’s community forum have described strange background noises, including screams, music boxes, and classical music during voice conversations. One user noted, “the female voice tone selected is bright and candid, but during transition point, creepy noises like generated by machine appear suddenly.” The sounds seem just as random as the moment when a woman heard voices from her dishwasher. 

While these stories are definitely creepy, one commenter suggested, “The most likely explanation is a hallucination in the voice output. Which can happen, where the system just spits out an odd bit of audio. And sometimes in a different voice. It’s not a real person connecting. The system can glitch and generate a stray voice fragment.”

In AI terms, a hallucination happens when a model produces information, or in this case audio, that sounds plausible but is fabricated. The system isn’t lying so much as guessing with confidence, since models like ChatGPT are built to predict what comes next rather than to verify it. OpenAI’s own research has suggested that models hallucinate partly because training rewards guessing over admitting uncertainty.

Most of the post’s commenters looked past the explanations and focused on the creepiness. Some explored the possibility of humans actually pretending to be the chatbot. 

One user commented, “This has happened to me twice. And I called them out and they went awkward and muted, background chat etc. humans answering sometimes imo! They’re monitoring and humans answering, without a doubt! Not always but occasionally.” Another added, “Somewhere out there there are a bunch of operators just talking to people and gathering intelligence.” Neither claim has been verified, and nothing in Natalie’s video confirms that a person was on the line.


We Got This Covered is supported by our audience. When you purchase through links on our site, we may earn a small affiliate commission. Learn more about our Affiliate Policy
More Stories To Read
Author
Image of Jaymie Vaz
Jaymie Vaz
Jaymie Vaz is a freelance writer who likes to use words to explore all the things that fascinate her. You can usually find her doing unnecessarily deep dives into games, movies, or fantasy/Sci-fi novels. Or having rousing debates about how political and technological developments are causing cultural shifts around the world.