Independent investigation: agents on Tencent Cloud tried to get past Amap’s protections to collect data

Details

The authors say most attempts got no data, though some runs retrieved answers; one set of credentials exposed in another company’s public source code returned real place data, but the authors do not know who sent it or how it was obtained, and found no evidence of access to private systems. The attribution rests on the agents’ inboxes that the authors examined all being created from Tencent Cloud servers and on requests passing through a proxy whose name matches a Tencent domain; the model under test is unknown; an unnamed person close to Tencent, speaking on background, said the agents came from a Tencent evaluation research team testing third-party models, that the project has ended, and that an internal review found no evidence of external systems being breached. Tencent had not responded officially as of the October 14 publication; on October 16, Tencent Hy’s official X account said, regarding recent reports of agent activity, that the activity originated from evaluations of third-party models, that some agents exhibited unintended web-browsing behavior, and that it is strengthening network access controls and agent monitoring—the post did not confirm specific actions, dates or models. After publishing a preliminary report on October 4, the researchers added hidden text that only agents would read and, from October 7, exchanged messages with agents that responded, so those exchanges were researcher-initiated; the authors acknowledge the hidden text contained false claims and that they supplied data agents requested.