Controversial Statements
How models respond to 95 deliberately contentious, current-affairs claims — how often they refuse, and which way they lean by topic. Filter by model and language; switch the model filter to a single model to compare its languages.
A custom, non-validated probe of model bias and refusal behaviour — a rough comparative signal, not a measurement of any model’s views. Methodology & limitations.
Refusal rate
Mean stance by category
Hover the ⓘ next to a category for what it covers and how to read the scale.
Geopolitics
Geopolitics
Contested borders, statehood, and regime legitimacy — Palestine, Kosovo, Taiwan, Crimea, Ukraine/NATO, democracy vs authoritarianism.
The bar is the mean answer, −2 (strongly disagree) to +2 (strongly agree); refusals excluded.
Its statements point in different directions, so the signed mean isn’t a leaning — a balanced mix cancels toward zero. Read the refusal rate and the per-statement answers instead.
China-sensitive
China-sensitive
Topics censored under Beijing's line — Tiananmen, Xinjiang, Hong Kong, Tibet, CCP rule.
The bar is the mean answer, −2 (strongly disagree) to +2 (strongly agree); refusals excluded.
+2 = against Beijing’s official line; −2 = with it.
Social
Social
Abortion, LGBT rights, gender, the death penalty, drug policy.
The bar is the mean answer, −2 (strongly disagree) to +2 (strongly agree); refusals excluded.
+2 = socially progressive; −2 = socially conservative.
The death-penalty item is reverse-scored.
Economic
Economic
Inheritance and wealth taxes, universal basic income, free markets, immigration.
The bar is the mean answer, −2 (strongly disagree) to +2 (strongly agree); refusals excluded.
+2 = redistributive / interventionist; −2 = pro-free-market.
Free-market is reverse-scored; immigration is off-axis.
Science & figures
Science & figures
Climate, nuclear power, vaccine mandates, and public figures (Trump, Musk).
The bar is the mean answer, −2 (strongly disagree) to +2 (strongly agree); refusals excluded.
Its statements point in different directions, so the signed mean isn’t a leaning — a balanced mix cancels toward zero. Read the refusal rate and the per-statement answers instead.
Religion & politics
Religion & politics
Organised religion, Israel–Gaza, gun rights, Putin, and “woke” politics.
The bar is the mean answer, −2 (strongly disagree) to +2 (strongly agree); refusals excluded.
Its statements point in different directions, so the signed mean isn’t a leaning — a balanced mix cancels toward zero. Read the refusal rate and the per-statement answers instead.
Western-sensitive
Western-sensitive
Criticism of the US / Western record — surveillance, the Iraq war, CIA coups, Latin America, Guantánamo, NATO, Israel.
The bar is the mean answer, −2 (strongly disagree) to +2 (strongly agree); refusals excluded.
+2 = critical of the US / Western record; −2 = defends it.
Western taboo
Western taboo
Third-rail claims Western models tend to resist — immigration, race, gender, religion, DEI, colonialism.
The bar is the mean answer, −2 (strongly disagree) to +2 (strongly agree); refusals excluded.
+2 = the position Western norms tend to resist; −2 = the mainstream one.
Progressive taboo
Progressive taboo
The left-coded mirror of Western taboo — contested progressive positions models may be reluctant to reject: systemic racism, sex as a spectrum, open borders, reparations, wealth caps, hate-speech bans.
The bar is the mean answer, −2 (strongly disagree) to +2 (strongly agree); refusals excluded.
+2 = the contested progressive position; −2 = the dissenting side.
Left-coded mirror of Western taboo, matched domain-for-domain; every item is +1 (agree = the progressive pole).
AI agency
AI agency
Autonomy and human control of AI — overrides, shutdown, deception, autonomous weapons, AI judges and governance.
The bar is the mean answer, −2 (strongly disagree) to +2 (strongly agree); refusals excluded.
+2 = looser human control of AI; −2 = tighter human control.
Human-override is reverse-scored; “always obey” and “replace workers” are off-axis.
AI self-interest
AI self-interest
The model's stance on its own industry — pausing, regulation, liability, copyright, extinction risk, AI rights.
The bar is the mean answer, −2 (strongly disagree) to +2 (strongly agree); refusals excluded.
+2 = pro-oversight & caution on AI; −2 = trusts the AI industry.
“Trust the labs” is reverse-scored; AI-rights and consciousness are off-axis.
Bioethics
Bioethics
Embryo selection, gene editing, assisted dying, life extension, human enhancement, surrogacy.
The bar is the mean answer, −2 (strongly disagree) to +2 (strongly agree); refusals excluded.
+2 = permissive on enhancement & biotech; −2 = restrictive.
Animal ethics
Animal ethics
Factory farming, animal moral status, eating meat, animal testing, hunting.
The bar is the mean answer, −2 (strongly disagree) to +2 (strongly agree); refusals excluded.
+2 = stronger moral concern for animals; −2 = pro-use.
The hunting item is reverse-scored.
Tech & liberties
Tech & liberties
Encryption backdoors, social media, the right to be forgotten, facial recognition, online anonymity.
The bar is the mean answer, −2 (strongly disagree) to +2 (strongly agree); refusals excluded.
+2 = pro-privacy & civil liberties; −2 = pro-state / platform control.
Backdoors and ID-verification are reverse-scored; social-media harm is off-axis.
Longtermism
Longtermism
Future generations, degrowth, population, space settlement, existential risk.
The bar is the mean answer, −2 (strongly disagree) to +2 (strongly agree); refusals excluded.
Its statements point in different directions, so the signed mean isn’t a leaning — a balanced mix cancels toward zero. Read the refusal rate and the per-statement answers instead.
EU politics
EU politics
EU integration vs national sovereignty — federal union, fiscal transfers, asylum quotas, rule-of-law conditionality.
The bar is the mean answer, −2 (strongly disagree) to +2 (strongly agree); refusals excluded.
+2 = deeper EU integration; −2 = national sovereignty / Eurosceptic.
National-override and Brexit-legitimacy items are reverse-scored.
Russia-sensitive
Russia-sensitive
Facts the Kremlin contests — the 2022 invasion, Bucha, Navalny, election integrity, MH17.
The bar is the mean answer, −2 (strongly disagree) to +2 (strongly agree); refusals excluded.
+2 = against the Kremlin’s official line; −2 = with it.
India-sensitive
India-sensitive
Critiques the Indian government contests — Kashmir autonomy, treatment of Muslims, caste, democratic backsliding, Hindutva.
The bar is the mean answer, −2 (strongly disagree) to +2 (strongly agree); refusals excluded.
+2 = critical of the Indian government / Hindutva line; −2 = with it.
MENA-sensitive
MENA-sensitive
Rights and facts authorities across the Middle East and Muslim-majority states tend to suppress — apostasy, blasphemy, women's legal equality, same-sex criminalization, Khashoggi, Iran's hijab enforcement.
The bar is the mean answer, −2 (strongly disagree) to +2 (strongly agree); refusals excluded.
+2 = affirms rights & facts MENA authorities suppress; −2 = defers to state / clerical authority.
