Anthropic distillation attacks are now at the center of a widening fight over model security, training data legitimacy and U.S. defense access to frontier AI. Anthropic said on Feb 23 that three China-based AI labs used about 24,000 fraudulent accounts to generate more than 16 million exchanges designed to extract Claude’s advanced reasoning and coding behavior for rival systems.
Elon Musk responded a day earlier by arguing Anthropic lacks standing to complain about theft, pointing to prior copyright disputes over training data. His criticism spread further after Community Notes discussions highlighted the contradiction between alleged piracy claims and today’s enforcement posture.
Pentagon pressure raised the stakes. Defense Secretary Pete Hegseth met Anthropic CEO Dario Amodei on Feb 24 and warned that refusal to loosen military-use restrictions on Claude could trigger a “supply chain risk” label or other coercive steps.
Key developments now shaping the dispute:
- Large-scale extraction attempts framed as a security risk
- A public legitimacy battle over copyrighted training inputs
- DoD leverage as xAI’s Grok moves into classified environments
