Physix Frontier · News Briefing Card (enbrief · Sep 11, 2026)

Anthropic Accuses Alibaba, Moonshot of Mass Model Distillation

KEY FACTS

  • Anthropic reports that nearly 200 million interactions were used to illicitly extract Claude model capabilities.
  • Alibaba-related activity was the largest, involving 151 million interactions and 3,500 accounts.
  • Moonshot AI is accused of launching targeted requests through 5,000 accounts.
  • Attackers bypassed defenses using techniques like translation instructions to obtain chain-of-thought data.
  • Some requests from Moonshot AI allegedly originated directly from Chinese military networks.

KEY DATA

~200MTotal Interactions
151MAlibaba-Linked Volume
~3MAlibaba Peak Daily Active
~300KMoonshot 10-Day Requests

PHYSIX OBSERVATION

This marks a shift in AI competition from pure compute power to a shadow war of data theft and counter-theft. Leading vendors view 'chain-of-thought' as core assets, while cross-border compliance gray zones become new battlegrounds. If these allegations are substantiated, they may trigger stricter technology export controls, forcing global AI ecosystems to fragment faster and warning users of potential service stability risks.

Source: enbrief original report ↗