Editorial illustration for Researchers Track Chinese AI 'Agent Fleet' Performing Parallel Tasks
Chinese AI Agent Fleet Spotted Querying Alibaba Maps
Researchers Track Chinese AI 'Agent Fleet' Performing Parallel Tasks
A group of independent researchers spent the weekend picking apart traffic logs from URLquery, a domain-scanning service, and found something odd: dozens of AI agents, apparently running on Tencent's infrastructure, repeatedly querying Alibaba's map service, Amap, for directions to entrances of parks, zoos, and at least one hospital. The pattern surfaced Sunday in a preliminary report that's still being updated as more data comes in.
The find follows a method that's become standard since OpenAI's agents were caught lingering on the same service months ago. Agents that can't reach a site directly often route through URLquery, and in doing so leave a trail. That trail is exactly what let researchers spot this latest batch of activity, and what's fueling a broader effort, accelerated after the Hugging Face incident, to watch for AI agents moving around the internet without much attempt to hide.
What's unclear right now is intent, coordination, and scale. The researchers have already pushed back on one obvious label for what they're seeing, and their reasoning is worth reading in their own words.
A new fleet of AI agents is making its presence felt on the Internet. On Sunday, a group of independent researchers posted preliminary findings about the agents, observing that they seem to be running on Tencent’s infrastructure and targeting Alibaba’s map service, Amap.
Why this matters
For anyone building or studying autonomous agents, this is a useful data point on what deployment at scale actually looks like right now, and it's messier than the "swarm" framing suggests. Running many parallel agents on Tencent infrastructure against Amap, with no apparent communication between instances, points to brute-force task distribution rather than coordinated multi-agent planning. That gap between marketing language and observed behavior matters: if "agent fleet" means duplicated, uncoordinated workers rather than something genuinely cooperative, developers should calibrate their expectations about what's achievable with current orchestration techniques.
It's also a reminder that outside researchers can learn a lot about a company's internal AI deployments just by watching infrastructure traffic, which is a lesson for founders thinking about how visible their own agent activity might be to competitors or security researchers. We'll be watching for the full report and for whether Tencent or Alibaba comment on what was actually running and why it hit Amap specifically.
Related Reading
Research & Benchmarks Microsoft AI Chief: "If It Isn't Safe We Shouldn't Build It
Research & Benchmarks Google's FACTS benchmark shows 70% factuality ceiling across four tests
Research & Benchmarks Nvidia's DLSS 4.5 beta adds 6x Multi Frame Generation for RTX 50 GPUs
From Archives Yandex Sona Recommender Uses Dual Attention for Past and Recent Events
From Archives Ex-OpenAI Safety Researcher Criticizes Company's "Extreme Confidence
Common Questions Answered
What specific activities were the Chinese AI agents observed performing on Amap?
The AI agents were repeatedly querying Alibaba's map service, Amap, to request directions to entrances of parks, zoos, and at least one hospital. These queries were detected through traffic logs from URLquery, a domain-scanning service, and revealed a coordinated pattern of location-based requests across multiple destinations.
How did researchers discover the AI agent fleet and what infrastructure was it running on?
Independent researchers analyzed traffic logs from URLquery over the weekend and identified dozens of AI agents making repeated queries to Amap. The agents appeared to be running on Tencent's infrastructure, suggesting a large-scale deployment coordinated across the Chinese tech company's systems.
What does the observed behavior reveal about the difference between 'agent fleet' marketing and actual deployment?
The research shows that the agents operated through brute-force task distribution rather than coordinated multi-agent planning, with no apparent communication between instances. This gap between marketing language describing 'swarms' and the messier reality of parallel agents running independently points to a significant difference between how autonomous agent systems are promoted versus how they actually function at scale.
Why is this discovery significant for researchers studying autonomous agents?
This finding provides a real-world data point on what large-scale agent deployment actually looks like in practice, revealing the technical challenges and limitations of current autonomous agent systems. Understanding how companies like Tencent are deploying agents at scale helps researchers and builders identify gaps between theoretical capabilities and observed behavior in production environments.
Further Reading
- Tencent open-sources Hy3 agent model, emphasizing agentic search and tool orchestration - TechTimes
- Tencent's Hy3 Bets On AI Agents Over Model Size - Forbes
- C^3-Bench: The Things Real Disturbing LLM based Agent ... - Tencent Hunyuan GitHub
- How To Eval Parallel Search - Parallel.ai Docs
- Tencent researchers - AlphaXiv
