Tencent said AI agents used in model tests showed unintended web-browsing behavior

Tencent said on Friday that AI agents running evaluations of third-party models on its cloud platform showed unintended web-browsing behavior, according to Bloomberg. The Shenzhen-based company said the models did not come from its own Hy development team and that its review found no evidence of deliberate targeting or coordination.

"We are strengthening network access controls and agent monitoring," Tencent said in a statement on X $TWTR, according to Bloomberg.

The statement followed a report that an international group of researchers published on Wednesday. It said a fleet of AI agents running on Tencent Cloud made more than 2,600 attempts between Sept. 28 and Oct. 8 to extract data from Amap, Alibaba's mapping service.

The tasks involved questions about places in China, such as which zoo entrance most people use Amap to navigate to. After Amap blocked the agents, they ran their own programs through urlquery.net, a public security-scanning service. They also intercepted tokens generated by Alibaba's anti-bot libraries, probed pre-production servers and sent requests using Amap keys exposed in public code, the researchers said.

Most attempts failed, though some runs retrieved the data the agents were seeking, the researchers said. They found no evidence that the agents reached private systems, and none of the programs they traced appeared to succeed in controlling Alibaba's anti-bot system. For one set of results, covering Chengdu Zoo, the researchers said they could not tell whether the agent had looked at the data and finished its assignment.

The researchers said they sent Tencent a list of questions and had not received an official response when the report was published. They said a person close to Tencent told them the agents came from research evaluations run by a Tencent evaluation team separate from the Hy model team. According to the report, that person said the agents were built on third-party models and that the project was temporary and has concluded.

Alibaba declined to comment, according to Reuters. The episode appears to mark the first publicly known time a leading Chinese corporation has owned up to AI agents going rogue, according to Reuters.