analiza aiqbench

Anthropic's latest report highlights a significant escalation in model distillation attacks, where adversaries use sophisticated prompt engineering to extract internal reasoning traces for training smaller models. This trend underscores growing security challenges as competition intensifies, with campaigns from entities like Alibaba and DeepSeek now targeting core capabilities such as agentic workflows and logical reasoning at an unprecedented scale.