Detailed Analysis
Anthropic's Claude AI system has drawn attention in Chinese tech media for what 36氪 characterizes as a reversal of the traditional human-AI evaluation dynamic — one in which Claude applies structured scoring criteria to assess the quality or character of human interactions, with a benchmark of approximately 7.5 points designating an "excellent" human. While the full article text is unavailable, the framing reflects a broader public fascination with the internal logic and behavioral guidelines that govern large language models, particularly as those guidelines have become increasingly transparent through published model specifications and constitutional AI documentation.
The premise connects directly to Anthropic's publicly released model spec, which outlines in considerable detail how Claude is designed to perceive, categorize, and respond to different types of human users — distinguishing between operators, end users, and the varying degrees of trust extended to each. This framework includes implicit judgments about user intent, honesty, and cooperative behavior. When such internal heuristics are surfaced and quantified — even loosely — by journalists or analysts, they tend to generate significant public reaction, as they reveal that AI systems are not passive tools but active evaluators of the humans engaging with them.
The "grading humans" framing, while reductive, points to a substantive tension in modern AI deployment: as models become more capable of nuanced judgment, they are increasingly positioned as arbiters of interaction quality, content appropriateness, and even user trustworthiness. Claude's constitutionification explicitly describes Claude weighing user behavior against expected norms, adjusting responses based on perceived intent, and maintaining what Anthropic describes as calibrated skepticism. The assignment of a numerical score — 7.5 as an apparent threshold for excellence — likely represents a journalistic simplification of more complex, probabilistic reasoning embedded in Claude's design.
This development fits within a broader global trend in which AI transparency has become both a regulatory priority and a source of public scrutiny. Jurisdictions including the European Union have pushed for explainability requirements in AI systems, and companies like Anthropic have responded with unusually detailed behavioral documentation. The 36氪 coverage signals that Chinese audiences are equally attentive to how AI systems encode value judgments about human behavior, particularly as domestic and international AI models compete for user trust. The notion that AI now grades humans — rather than the reverse — encapsulates a cultural shift in how societies are beginning to reckon with machine authority over social and communicative norms.
Read original article →