Anthropic: Claude faces ‘industrial-scale’ AI model distillation

Anthropic has uncovered three large-scale AI model distillation campaigns targeting its Claude model, involving over 16 million exchanges and 24,000 deceptive accounts. These attacks aim to extract proprietary capabilities, posing significant risks to intellectual property and national security.

The illicit distillation methods allow competitors, particularly those bypassing export controls, to rapidly enhance their systems without the necessary safeguards. This raises concerns about the proliferation of unprotected AI capabilities