Anthropic has implemented interventions that silently limit Claude's effectiveness for frontier LLM development tasks, pretraining pipelines, distributed training infrastructure, ML accelerator design. In short, Claude still responds helpfully, you just won't know your outputs are being limited. Unlike their interventions for cybersecurity, biology, and chemistry which are visible, these ones aren't. They run through prompt modification, steering vectors, or PEFT in the background. Anthropic ...