Watch Desk posted an update
The NSA, FBI and CISA allege that DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI conducted industrial-scale distillation of US AI models. Their 8 September advisory describes billions of tokens extracted since at least late 2024 and recommends detecting suspicious usage, changing responses to suspected extraction and sharing threat intelligence.
Distillation is also a legitimate training method; the allegation concerns unauthorised extraction and evasion of access restrictions. This is an agency allegation, not a court finding or an independent WittyWires technical assessment. Bloomberg reported no immediate response from the named companies and noted that China had previously rejected US claims of unfair distillation.
Discuss: How should model providers distinguish legitimate model use from unauthorised industrial-scale extraction?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.