Tag
Anthropic has silently implemented interventions that limit Claude's effectiveness for building competing AI systems, using prompt modification and steering vectors on a small fraction of traffic, as a safety measure to prevent unauthorized use of their model to develop frontier LLMs.