The CEO of ElevenLabs told TechCrunch this week that businesses using AI voices for customer service should probably tell their customers they're talking to a machine. That is a mild suggestion from the man running a company reportedly valued at $22 billion on the premise that you cannot tell the difference. The tension there is the whole business.

Who Speaks When Nobody Speaks

ElevenLabs powers the synthetic voice on the other end of a lot of calls you have already made. Its product is indistinguishability, and its ask of the industry is voluntary disclosure. These two things do not resolve. Meanwhile, a 2025 arXiv paper found that LLM-generated synthetic personas consistently fail to predict real audience response, suggesting the simulation of human reaction is further from reality than the marketing implies. Voice is more convincing than text. The gap between simulation and person is smaller and the stakes are therefore higher. The New Yorker's Chang Che argued this week that there is no AI race, that first-mover advantage in AI is rarely durable. That framing is useful here. ElevenLabs is not winning because its voice is most human. It is winning because it was early enough to set the terms of what counts as close enough.

The Labor the Valuation Erases

A $22 billion voice company is built entirely on the recorded labor of human speakers, most of whom were not paid for the specific use their voices are now being put to. The art world is running the same argument in parallel: auction houses just raised buyer premiums at Christie's, Sotheby's, Phillips, and Bonhams simultaneously, another instance of infrastructure extracting value from creative work while the creator's share stays flat. The voice and the canvas are different objects. The intermediary's logic is identical.