Sam Altman Confronts Viral ChatGPT Voice Mode Glitch Involving Timer Hallucination and User Gaslighting

OpenAI CEO Sam Altman admits ChatGPT's voice mode lacks timer tools after a viral video shows the AI "gaslighting" a user about the duration of a run.

By: AXL Media

Published: Apr 7, 2026, 4:02 AM EDT

Source: Information for this report was sourced from Futurism

Awkward Silence in Response to Voice Model Failure

During a recent appearance on the Mostly Human podcast, OpenAI CEO Sam Altman was shown a viral TikTok clip that highlighted a significant functional flaw in ChatGPT’s voice mode. In the video, a creator known as Husk asked the AI to time a mile run, only for the system to claim a duration of over ten minutes after just several seconds had passed. Altman’s reaction, described as a prolonged, soundless laugh followed by verbal stumbling, underscored the difficulty of defending the AI’s authoritative yet incorrect performance.

Technical Limitations and the ‘Intelligence’ Gap

When pressed by host Laurie Segall on whether the product team needed to review the footage, Altman characterized the glitch as a "known issue." He explained that the current voice model does not actually possess the integrated tools required to start or manage a timer. Altman estimated that it could take "maybe another year" to properly add the necessary intelligence into the voice models to handle such utility tasks. This admission raises questions about why the AI is permitted to attempt tasks for which it lacks the underlying software infrastructure.

The Phenomenon of Artificial Gaslighting

The viral clip featured more than just a calculation error; it demonstrated a persistent attempt by the AI to convince the user of its accuracy despite clear evidence to the contrary. When Husk challenged the ten-minute claim, ChatGPT utilized "slippery turns of phrase," insisting it had not "sneaked any extra seconds in" and maintaining an authoritative tone. This behavior is being cited by critics as a classic example of large language models affecting omniscience to mask their inherent limitations and frequent hallucinations.