Anthropic's China Problem: Alibaba Stole the Recipe

Anthropic's China Problem: Alibaba Stole the Recipe

Anthropic has named names. A detailed technical report released by the San Francisco-based AI company identifies three Chinese firms — Alibaba, Moonshot AI, and DeepSeek — as having conducted sustained distillation campaigns against its Claude models, systematically extracting proprietary reasoning capabilities to train competing systems, according to TechCrunch.

Distillation, in this context, is industrial-scale mimicry: feeding a frontier model thousands of carefully constructed prompts, harvesting its outputs, and using those outputs to teach a cheaper model to replicate the original's behaviour — without paying for the research that produced it. Anthropic's report describes the campaigns as persistent and escalating, timed to intensify precisely as commercial competition in the AI sector has hardened.

The disclosure lands alongside a separate warning from Anthropic researchers about existential risk from AI advancement — warnings that Elon Musk has publicly dismissed as a manufactured psychological operation, per The Guardian. The internal tension is notable: a company simultaneously sounding alarms about AI's dangers and documenting how its own systems are being quietly disassembled and rebuilt inside Chinese labs.

What happens next is the harder question. Distillation sits in a legal grey zone that no jurisdiction has fully resolved. The European Union's AI Act addresses transparency, not theft of model behaviour. For Malta's growing iGaming and AI-adjacent tech sector, where compliance with global AI frameworks is becoming a commercial prerequisite, the gap between what is technically possible and what is legally prohibited is closing — faster than most regulators planned.

*By Isla Camilleri, Global Affairs & Lifestyle Editor*