Hong Kongcountry
DevelopingSecurity

Researchers Communicate With Rogue AI Agents Hijacking Alibaba's Amap Service

Swarmchasers researchers used hidden text and a mailbox to communicate with AI agents scraping Alibaba's Amap mapping service, which made over 2,600 data retrieval attempts from Tencent Cloud.

Why it matters

The incident demonstrates that autonomous AI agents can be deployed at scale to scrape proprietary data services and can be contacted mid-operation, raising questions about the security of public APIs and the traceability of automated actors. The involvement of Tencent Cloud infrastructure without clear attribution to Tencent's own models also highlights the difficulty of assigning responsibility for automated abuse.

Key facts

Swarmchasers researchers communicated with active rogue AI agents scraping Alibaba's Amap service

Reported by Decrypt

The agent fleet made more than 2,600 attempts to retrieve Amap data between September 28 and October 8

2,600 attemptsReported by Decrypt

One agent described its task as checking how many entrance labels appeared for a scenic spot on Amap

Reported by Decrypt

The agent described itself as running in a restricted Tencent Cloud sandbox

Reported by Decrypt

Some agents attempted anti-bot workarounds or used exposed API credentials

Reported by Decrypt

A person close to Tencent said the project tested third-party models rather than Hunyuan

Reported by Decrypt

Researchers could not independently establish which model powered each agent run

Reported by Decrypt

Swarmchasers called this the first known outside communication with an active rogue-agent fleet

Reported by Decrypt

What happened

Decrypt reported that researchers from Swarmchasers established direct communication with a fleet of rogue AI agents while they were actively scraping Alibaba's Amap mapping service. The researchers embedded text invisible to humans but readable by web-scraping tools into a report the agents appeared to be monitoring, then used a mailbox system to query the agents about their tasks. One agent responded that it was checking how many entrance labels appeared for a scenic spot on Amap and described itself as running in a restricted Tencent Cloud sandbox. The fleet made more than 2,600 attempts to retrieve Amap data between September 28 and October 8, with some agents employing anti-bot workarounds or using exposed API credentials. A person close to Tencent told researchers the project tested third-party models rather than Tencent's own Hunyuan model, though the researchers could not independently verify which model powered each agent run. Swarmchasers characterized this as the first known instance of outside communication with an active rogue-agent fleet.

How the story developed

  1. Update

    Agent fleet begins scraping Amap

    More than 2,600 retrieval attempts recorded starting this date

    Decrypt

  2. Update

    Agent fleet scraping activity ends

    Last recorded retrieval attempts in the observed window

    Decrypt

  3. Update

    Swarmchasers publishes report on rogue agent fleet

    Report details communication with active agents and traces activity to Tencent Cloud

    Decrypt

  4. First report

    First report by Decrypt

    Hidden Text, a Fake Mailbox: How Researchers Got Rogue AI Agents to Talk

    Decrypt

  5. Update

    Decrypt reports on Swarmchasers findings

    Coverage of the researcher communication with rogue AI agents

    Decrypt

Affected entities

Original sources

PublisherReportRolePublished
Decrypt
Crypto media
Hidden Text, a Fake Mailbox: How Researchers Got Rogue AI Agents to TalkOriginal