Stories
Anthropic Cuts Claude AI Distillation Attempts by Up to 90 Percent
F
Firstpost•13-10-2026, 09:52
Anthropic Cuts Claude AI Distillation Attempts by Up to 90 Percent
- •Anthropic has introduced new safeguards that successfully reduced unauthorized AI model distillation attempts by 90 to 99 percent.
- •The company discovered that Chinese AI models, including DeepSeek and Moonshot, unlawfully used Claude to answer user queries.
- •Anthropic terminated tens of thousands of accounts and tightened controls against proxy networks to prevent capability extraction.
- •The new measures also identified suspected plots involving attempts to direct Claude toward developing biological weapons.
Key Questions Answered
Anthropic introduced new safeguards to restrict user access, tightened controls against proxy networks, and terminated tens of thousands of accounts linked to suspected distillation campaigns, successfully cutting reasoning data extraction by up to 90 to 99 percent.
Anthropic identified Chinese AI developer Z.ai, also known as Zhipu, as an organization that attempted to extract capabilities from Claude while developing its GLM-5.3 model.
Anthropic reported that it disrupted several plots where researchers tried to steer its models toward biological weapons development, adopting a cautious approach while keeping the system's false-positive rate below 0.1 percent.
