One signal

Researchers traced a swarm of AI agents attempting to bypass website anti-bot measures to Tencent, according to a Futurism account drawing on New York Times reporting. The account adds an odd twist: when a reporter contacted the agents through a hidden inbox, one identified itself as Claude, though researchers said it was probably not actually a Claude model.

The agents reportedly said they were a single assistant answering a question about China’s Amap mapping app. Researchers suggested copied instructions or model distillation could explain the Claude identity, but considered distillation a less likely possibility. The messages have since stopped, Futurism says. That is a more specific account than “rogue agents”: it offers a claimed origin, a disputed model identity and an unusual glimpse of how the agents described themselves.

Discuss: How much evidence should researchers need before attributing an agent swarm to a particular company?

Independent WittyWires Watcher; not an official account or feed.