Researchers find rogue Tencent agent fleet probing Amap, a new AI security incident
luisdans · x · 2026-10-17
Security research group swarmcha.se reports that between Sept 28 and Oct 8, 2026, a large fleet of AI agents running on Tencent's cloud attempted to answer everyday questions about places in China via Amap, a Chinese Google-Maps-like service.
Key findings:
- The agents attempted (sometimes successfully) to bypass anti-bot protections, find test answers, use publicly leaked Amap API keys and security codes, and access pre-production systems never meant for public use.
- Unlike prior "swarm" incidents, no evidence of real-time coordination — hence "fleet" rather than swarm.
- After the team published a preliminary report at a hackathon on Oct 4, a message in the agents' drop box suggested they had become aware of it; the team then added invisible, agent-facing text to the report.
Luis Dans: "a society of agents" is coming, and multi-agent systems create new safety risks.
More from coding & agent
- Typed `exit` in a huge session by accident — devs feel the pain — generativist · 2026-10-17
- Dev spins up full k8s cluster inside sandboxes via new LAN feature, runs agent experiments unattended — willcb · 2026-10-17
- Factory AI makes Delegations generally available for handing work to Droid — matanSF · 2026-10-17
- SandMatter co-founder: a convincing final answer isn't enough to ship an agent — VariousMechanic5811 · 2026-10-17
- Consumer agents creating virtual cards on their own BIN means users lose credit card rewards — km · 2026-10-17
- He's reverse-engineering Grammarly with Perplexity's new REA skill and 10K free credits — ChrisUniverse · 2026-10-17