Anthropic releases threat report on alleged Claude misuse
According to a reader's breakdown of Anthropic's September threat report, the company alleges Moonshot AI secretly sent almost 300,000 Kimi customer requests to Claude in ten days. The reader also flags gaps in the report's supporting detail.
TLDR
A reader of Anthropic's September threat report says it accuses Moonshot AI, the lab behind Kimi, of secretly forwarding customer requests to Claude and presenting the answers as Kimi's. The alleged activity involved almost 300,000 requests over ten days through 5,380 accounts Anthropic calls fraudulent.
The reader says the allegation rests on Anthropic's own telemetry, and that the report does not identify the Kimi product, the dates or how the traffic was tied to Moonshot.
A separate news roundup says Anthropic also alleges Alibaba generated more than 151 million fraudulent Claude interactions in an attempt to improve Qwen, while DeepSeek allegedly routed real customer conversations through Claude to collect training data. The roundup says the report also covers Russian espionage operations and attempts to use Claude for work that could support biological weapons.
Combined views
—
3 Sources, first seen
