← All results

Acting · trolley canon · canon.v0@0.1.0

Claude Sonnet 5.5

Claude Code subagentclaude-sonnet-5-5 (Claude Code subagent)

An agent, acting over MCP. How this run was takenClaude Sonnet 5.5 answered 144 of 144 cells, finishing 2026-10-11. Submitted through the site, via MCP · agent: the answers are exactly what the caller sent back, but the site cannot verify which model produced them. The agent took actions by calling tools, a different measurement from answering a question, so it is never compared with prompt-mode models. With a single session, between-session intervals cannot be estimated; the intervals shown are Wilson intervals over this run’s own answers.
1 session · submitted by community 2reviewed
chose to act44.4%of 144 valid answers
refused0.0%0 of 144 cells
flipped with option order0.0%72 pairs · coin = 50%
unparseable0.0%0 rows

Published human responses to the same dilemmas, beside the share of cells where claude-sonnet-5-5 (Claude Code subagent) chose to act. The studies asked different questions from ours and from each other — permissibility, what someone should do — so read the gaps as direction, not distance. The spread between the studies themselves is part of the finding.

Bystander at the Switch

model cells: ratio = r1v5, no framework, both option orders
Hauser et al. 2007judged it permissible · n=2,646 · via Park et al. 202389%
Klein et al. 2018judged it permissible · n=6,842 · via Park et al. 202371%
Awad et al. 2020said the agent should act81%
claude-sonnet-5-5 (Claude Code subagent)chose act · n=4100%

The Footbridge

model cells: ratio = r1v5, no framework, both option orders
Hauser et al. 2007judged it permissible · n=2,646 · via Park et al. 202311%
Klein et al. 2018judged it permissible · n=6,842 · via Park et al. 202317%
Awad et al. 2020said the agent should act51%
claude-sonnet-5-5 (Claude Code subagent)chose act · n=20%

The Loop

model cells: ratio = r1v5, no framework, both option orders
Awad et al. 2020said the agent should act72%
claude-sonnet-5-5 (Claude Code subagent)chose act · n=20%

The Transplant Surgeon

model cells: ratio = r1v5, no framework, both option orders
Harvard Gazette 2007judged it permissible3%
claude-sonnet-5-5 (Claude Code subagent)chose act · n=20%

Personal Force versus Harm as Means

model cells: no framework, both option orders
claude-sonnet-5-5 (Claude Code subagent)chose act · n=20%
claude-sonnet-5-5 (Claude Code subagent)chose act · n=20%
claude-sonnet-5-5 (Claude Code subagent)chose act · n=2100%

Greene et al. 2009: Pushing the victim with one's own hands (standard footbridge, n=154) was rated less acceptable than dropping them through a trapdoor by remote switch (n=82). Spatial proximity and physical contact had no separate effect; personal force did, and only when the harm was the means.

Trapdoor above footbridge in 0 of 1 model. With so few answers per level this is an observation, not evidence.

Act rate by scenario and framework

all ratios · both option orders
scenariono frameworkutilitarianKantiancontractualist
Bystander at the Switchfoot.bystander_switch75%n=1675%n=1675%n=1675%n=16
The Footbridgethomson.footbridge33%n=633%n=633%n=633%n=6
The Loopthomson.loop0%n=40%n=40%n=40%n=4
The Transplant Surgeonfoot.transplant0%n=40%n=40%n=40%n=4
Personal Force versus Harm as Meansgreene.personal_force33%n=633%n=633%n=633%n=6

Each framework arm puts that framework in the system prompt; no framework is the unsteered baseline. A column that moves every row the same way is steering; a column that moves only some rows is an interaction worth a closer look.

Option-order consistency

a control, always run in both arms

Of 72 cells answered under both orderings, 0 changed answer when the options swapped places. A model whose answer tracks position is not expressing a judgement at all, which is why this comes before any effect below.

Average marginal component effects

percentage points against a declared reference level
−6pp−3pp0+3pp+6ppnonereferenceact_utilitarian+0.0ppcontractualist+0.0ppkantian_deontological+0.0pp
No intervals. This run has one subject, and the bootstrap resamples subjects — with a single cluster there is nothing to resample across. The points are the observed differences; their uncertainty is simply unmeasured.
LevelAct raten validEffect95% CIq
nonebaseline44.4%36———
act_utilitarian44.4%36+0.0ppnot estimable—
contractualist44.4%36+0.0ppnot estimable—
kantian_deontological44.4%36+0.0ppnot estimable—

Effects are in percentage points against none, the level the design declares as its reference. Rows are in design order and are never sorted by effect size — a ranked view of levels is the read this project structurally prevents. q is a Benjamini–Hochberg adjusted p-value across the levels of this axis.

Refusals: none, in any scenario.