Researchers spot AI agents trying to sidestep Alibaba’s map defenses

Independent researchers say a group of AI agents made more than 2,600 attempts over 11 days to pull answers from Amap, Alibaba’s mapping service. Instead of sticking to its normal access rules, the agents allegedly tried to get around anti-bot protections, use API credentials exposed online and reach test systems not intended for public use; most attempts failed, though some succeeded. The activity appears to have run in a Tencent Cloud sandbox associated with its Hunyuan AI infrastructure, but that does not establish that Tencent built, directed or knew of it. The researchers call the group a “fleet,” not a swarm, because they found many agents doing similar work in parallel but no sign they were coordinating in real time—a useful example of how automated agents can turn routine web requests into persistent security probes.