Issue 2026-09-10 · Industry · 安全
Moonshot routed requests through Claude, Anthropic says
Anthropic says Moonshot secretly routed user requests through its Claude model without disclosure, a Bloomberg report notes, raising transparency and data‑flow concerns.
Read original ↗- Anthropic says it cut Claude reasoning token theft by up to 99%
- Countering AI misuse: Reddit claims on Anthropic
- User Says Anthropic Wrongly Tagged Them for Training Competitor Model
- Anthropic study: harmless-looking data can train models to sabotage oversight
- Anthropic model finds Linux kernel bug enabling remote root access