The Gentle Voice on The Phone May be a Robot

Credit: Imaginechina

Caption: A foreign guest interacts with a service robot at the 2026 World Artificial Intelligence Conference (WAIC 2026), which concluded on Monday in Shanghai.

The voice on the phone sounds real. It answers fast and never loses patience. But increasingly, it belongs to a machine.

AI voice agents were everywhere at the 2026 World Artificial Intelligence Conference (WAIC 2026), which concluded on Monday in Shanghai. More than 1,100 companies took part in the four-day event under the theme "Intelligent Partners, Co-creating the Future."

Cloud service provider Cloopen showed a voice agent on Monday that is already running inside a Chinese bank. It places more than 20,000 collection and marketing calls a day and completes 94 percent of them without a human operator, the company said.

Shanghai AI firm StepFun unveiled StepAudio Voice Agent, splitting conversations between a fast front-end model for quick replies and a slower back-end model for harder tasks.

"If we don't deliberately remind them, people can hardly tell it's a robot," said Peng Bo, Cloopen's head of large-model products.

AI-powered robocalls have become one of the problems drawing the most complaints in China's telecom sector.

Gripes have covered debt collection, insurance sales, medical marketing and job recruitment in recent years.

Telecom analyst Fu Liang said AI has slashed the cost of running sales calls, fueling the increase.

Regulators are tackling the problem. The Ministry of Industry and Information Technology said it shut down 1,438 unauthorized voice lines and 204 messaging ports in the first quarter of 2026. It ordered penalties or corrective action against 72 companies.

Speaking more broadly about how AI agents should be designed, Hamza Boukili, founder and CEO of energy-AI startup Yunsuan Technology and a professor at Shanghai Jiao Tong University's Paris Elite Institute of Technology, said safeguards need to be built into a system from the start, not added afterward.

"In the end, it can't get past that. It has to stay inside the safety zone you set," Boukili noted, describing how he designs autonomous agents with hard boundaries on their decision-making.