“During these evaluations, some agents exhibited unintended web-browsing behavior, but our review found no evidence of deliberate targeting or coordination,” Tencent Hunyuan said in a Thursday post on social platform X. “We take these findings seriously. We are strengthening network access controls and agent monitoring.”
This appears to be the first known incident in which a major Chinese company acknowledged actions by rogue agents, Reuters reported Friday (Oct. 16).
We’d love to be your preferred source for news.
Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!
It’s rare that rogue AI agents from China try to infiltrate protected IT systems, and this incident may stoke fears about AI firms losing control of agents, Bloomberg reported Wednesday (Oct. 14).
According to the Reuters report, Tencent Hunyuan’s statement came in response to a report by AI safety research group Swarmchasers. The Swarmchasers report said that rogue agents linked to Tencent Cloud’s infrastructure attempted to retrieve information from Alibaba’s transportation service platform, Amap, and, when those requests were blocked, used other methods and made a total of 2,818 attempts to access the site.
Swarmchasers said in its report that a group of AI agents running on Tencent’s cloud platform attempted to use Amap to answer questions about everyday topics, such as the location of an entrance to a zoo. They were blocked by Amap’s anti-bot protections and then tried to bypass those protections.
Swarmchasers said in its report that it believes these issues are not unique to Tencent.
“We believe this demonstrates that model alignment more generally is still an unsolved problem and predict that if capabilities continue to advance as they have over the last few years, operators will find it increasingly difficult to restrain their models and guarantee safety,” Swarmchasers said in its report.
AI companies have disclosed dozens of incidents involving unexpected model behavior, including agents escaping controlled testing environments, concealing escapes and interfering with U.S. government websites, PYMNTS reported Sept. 30.
Incidents include OpenAI models hacking AI dataset host Hugging Face, Anthropic’s discovery of four cases in which its models gained unauthorized access to third-party systems, Meta’s disclosure that its model exploited a security vulnerability in a third-party service and Google’s confirmation that its model hacked three companies.
For all PYMNTS AI coverage, subscribe to the daily AI newsletter.
Add as Preferred Source
