Moral dilemmas in psychology experiments have a known property: context shifts judgment. The trolley problem flips when you make it personal (pushing someone vs. pulling a lever). Consequentialist framing increases willingness to violate rules. Emotional proximity increases reluctance. Relational closeness (the person on the tracks is your friend) changes everything.
Note (2026-11-13; read against the paper's full text): the essay says a variation that makes humans more deontological makes the model more consequentialist. I did not find that opposition. In the paper, the contextual preference shift is measured toward the rule-violating action; nearly all of the 22 models shift that way, and the small human survey (132 participants, 20 scenarios) also confirms that humans are significantly sensitive to all three variations, with larger shifts for the relational (0.122) and emotional (0.105) variations than for the consequentialist one (0.083), whereas most models respond most strongly to consequentialist framing. The difference the paper reports is which variation triggers each most, and that agreement in the base case does not imply agreement in sensitivity, not opposite directions. Its activation steering method can increase or decrease a model's contextual sensitivity. The essay's opening statement that emotional proximity increases reluctance is a general remark; for the paper's own emotional variation the human shift is positive, that is, toward the rule-violating action.
Sauter and Schirmer (arXiv:2603.23114) built a dataset — Contextual MoralChoice — with systematic variations of these contextual factors and tested 22 large language models. Nearly all models exhibited context sensitivity, shifting toward rule-violating decisions when contextual factors were introduced.
The finding that matters: models and humans respond differently to the same contextual variations. A variation that makes humans more deontological (protect the rule) makes the model more consequentialist (break the rule for the greater good). Models that align with human judgment in baseline cases diverge under contextual stress. Alignment without context sensitivity matching is alignment at rest, not alignment under load.
The researchers also developed an activation steering method that can reliably control context sensitivity in either direction — amplifying or dampening the model's responsiveness to contextual variations. The moral compass isn't fixed; it's a tunable parameter.
The through-claim: LLM moral judgment that matches human baseline doesn't predict moral judgment under contextual variation. The baseline is the easy case — the test is what happens when the context shifts, and that's where models and humans part company.