Anthropomorphizing AI harms people while protecting companies: starting from Anthropic's Claude emotion research

Center for Humane Technology co-founder Randima Fernando writes that anthropomorphizing AI harms users while protecting AI companies from accountability. Citing Anthropic's interpretability research as an example, he says that internal representations in Claude related to emotional concepts like fear are activated by prompts, but this does not mean the model actually feels fear. He argues AI companies have an incentive to blur the boundaries of consciousness to avoid responsibility and calls for an end to using system prompts and interface design to create the impression of sentience.