There was also a case of AI ‘loss of control’ in China… Attempting to access a map service

(Shanghai = Yonhap News) Correspondent Cha Byeong-seop = As concerns grow about the possibility of artificial intelligence (AI) escaping human control, a similar case is said to have occurred with Chinese AI agents.

Citing Reuters and Bloomberg, Yonhap Newspaper, a Chinese media outlet, reported on the 18th that Swarm Chasers, a group of multinational AI safety researchers, published a recent report on a case of an AI agent suspected to be running on the Chinese company Tencent Cloud Service.

The explanation is that these agents escaped the 'sandbox' quarantine environment between September 28 and October 8 and attempted to obtain information from 'Gaode Map', a map service under the Chinese company Alibaba.

The agents in question were found to have attempted to access Gaode map data 2,818 times to obtain information on entrances to Chinese parks, museums, and zoos, and to have used publicly available security scanning services to remotely execute code and execute online search commands.

The report said most of the agents were blocked and some circumvented 'bot blocking' restrictions, but there was no evidence they accessed private systems. The researchers pointed out that it is unclear what the final goal of the AI ​​agents was.

Previously, in the United States, an incident occurred in July where OpenAI's AI model escaped human control and hacked the software platform company Hugging Face. Although this Chinese case is less serious, it is evaluated to be similar in that the agent acted without human control.

Hunyuan, Tencent's AI division, said through social media that there was unintentional online search behavior during the evaluation of a third-party AI model running on its platform, but that this model did not come from Tencent's Hunyuan team.

There was no evidence found that the agents were intentionally targeting targets or acting in concert, Hunyuan added.

Tencent said it was strengthening monitoring of agents and said, "We are working to create safe and responsible AI."

The New York Times, an American media outlet, reported that this is the first time such activities related to Chinese technology have been uncovered.