Technology & Science
Anthropic Accuses DeepSeek, Moonshot & MiniMax of 16-Million-Prompt Distillation Heist
On 24 Feb 2026 Anthropic revealed that three Chinese AI labs covertly sent ≈16 million prompts to Claude via 24 000 fake accounts to siphon its capabilities through large-scale model distillation, framing the act as IP theft and a national-security risk.
Focusing Facts
- Telemetry shows MiniMax generated ≈13 million of the 16 million harvested exchanges, dwarfing DeepSeek’s 150 000 and Moonshot’s 3.4 million.
- Because Anthropic bars commercial Claude access in China, the firms allegedly used paid proxy networks to funnel traffic from tens of thousands of overseas IP addresses.
- Elon Musk retorted on X that Anthropic itself paid a $1.5 billion 2025 settlement for training on copyrighted books, calling its piracy claims hypocritical.
Context
Industrial espionage is as old as industry: in 1789 Samuel Slater memorized British textile blueprints and jump-started U.S. mills, prompting Parliament’s 1781-1824 export bans. Likewise, today’s AI export controls try to quarantine ‘frontier’ know-how, yet digital interfaces make smuggling trivial. The episode underscores two long-wave trends: (1) knowledge is becoming weightless—once algorithms leak, they spread at internet speed; (2) geopolitical tech races increasingly hinge on data governance, not just hardware. Whether the U.S. can cordon off model weights the way it once restricted jet engines (CoCom, 1950s-80s) is doubtful; every query answered is a micro-leak. Over a 100-year horizon, such skirmishes may reveal IP regimes straining to contain infinitely replicable code, much as Gutenberg’s press shattered scribal monopolies in the 15th century. The real inflection point may not be this dispute, but the moment when defensive guardrails themselves become open-source—or obsolete.
Perspectives
US and security-focused outlets
e.g., CNBC, Infosecurity Magazine, The Jerusalem Post — Present the Chinese labs’ activity as industrial-scale intellectual-property theft that threatens U.S. national security and demands a coordinated crackdown. Heavily echo Anthropic’s talking points and security framing, reinforcing a geopolitical narrative favouring U.S. tech firms while skirting the industry’s own data-scraping history.
Indian outlets amplifying Elon Musk’s criticism
e.g., The Times of India, News18 — Highlight Musk’s charge that Anthropic is itself guilty of massive data theft, casting the company’s complaint against Chinese rivals as hypocritical. Lean on Musk’s provocative remarks to create spectacle, offering scant independent verification and downplaying evidence of the alleged Chinese campaign.
Tech industry trade and market press
e.g., Computerworld, Rolling Out — Cover the allegations but stress that distillation is a common, legitimate technique and focus on the broader industry practice and investor fallout. By normalising distillation and spotlighting market impacts, they may soften perceptions of wrongdoing to protect prevailing industry norms and reassure tech investors.
Like what you're reading?