Anthropic accuses Alibaba, Moonshot AI and DeepSeek of massive model distillation

The distillation fight between Anthropic and Chinese labs makes explicit what many in the industry assumed: once a few actors reach the frontier, everyone else can try to compress their capabilities rather than reinvent them. If Alibaba, Moonshot AI and DeepSeek can cheaply harvest Claude outputs at industrial scale, they can shorten their own training cycles and reduce the capital required to produce competitive models. That raises the floor of what ‘frontier’ capability means and increases the number of serious players in the game.

At the same time, public accusations like this push model providers toward more aggressive API hardening, regional blocks and legal action. Combined with the recent NSA and FBI advisory on Chinese distillation campaigns, the picture that emerges is an AI race that looks less like open scientific competition and more like a contested strategic technology, where cross border cloning attempts and countermeasures are the norm. In that environment, each increment in model capability is likely to spread faster via distillation than via original research, tightening timelines even if a few Western labs try to self regulate.