Claude Maker Says Safeguards Cut Illicit Reasoning Extraction by 10 to 100 Times. Anthropic said Tuesday it developed new methods to styme competitors from distilling its reasoning models, an ongoing problem for which the company has asked for government support. Anthropic says it will limit the portability of thinking blocks from one model to another.
First seen on govinfosecurity.com
Jump to article: www.govinfosecurity.com/anthropic-rolls-out-defenses-to-block-ai-model-distillation-a-33093
