Preferring to completely eliminate a small risk rather than making a larger reduction in overall risk that would help more.
Imagine you have 10 band-aids. You have a tiny scratch on your finger and a big cut on your knee. Instead of putting most band-aids on the big cut to stop the bleeding, you use all 10 on the tiny scratch so it's perfectly covered with zero chance of getting dirty — while the big cut just keeps bleeding. You feel better because one boo-boo is totally fixed, but overall you're much worse off.
Zero-Risk Bias describes a systematic preference for options that completely eliminate one identifiable risk, even when an alternative option would produce a greater total reduction in danger across all risks combined. This bias leads people to place disproportionate psychological value on achieving absolute zero in a single risk category rather than optimizing for the best aggregate outcome. It is especially pronounced in emotionally charged domains like health, safety, and environmental policy, where the symbolic reassurance of total elimination outweighs rational cost-benefit calculations. The bias helps explain why individuals and policymakers routinely funnel limited resources toward eliminating negligible threats while leaving far larger hazards inadequately addressed.
The same glitch looks different depending on the terrain. Finance, medicine, a relationship, a team — same mechanism, different costume.
Investors during periods of economic uncertainty tend to flock to guaranteed low-yield government bonds rather than diversified portfolios that would reduce overall financial risk more effectively, because the guaranteed return eliminates one specific type of loss entirely even though overall wealth-building is sacrificed.
Patients and policymakers may push for total bans on substances with negligible carcinogenic risk while underfunding interventions against far more prevalent killers like heart disease or diabetes, because eliminating even a tiny cancer risk feels morally absolute in a way that 'reducing' a larger risk does not.
Formalized by Viscusi, Magat, and Huber (1987) who first documented willingness to pay disproportionately to eliminate small risks, and further developed by Jonathan Baron, Rajeev Gowda, and Howard Kunreuther (1993) through hazardous waste cleanup survey experiments. The concept builds on the certainty effect described by Kahneman and Tversky's Prospect Theory (1979).
In ancestral environments, threats were discrete and localized — a snake in the path, a contaminated water source, a rival at the cave entrance. Completely eliminating a known, identifiable threat (killing the snake, avoiding the water source entirely) was a reliable survival strategy when threats were few and resources for managing them were limited. The brain evolved to reward total threat neutralization with strong feelings of safety and closure, because in small-scale environments, that was usually the optimal response.
AI safety and alignment research can exhibit zero-risk bias when teams focus disproportionate resources on fully preventing one narrowly-defined failure mode (such as a specific type of prompt injection) while leaving broader, more impactful vulnerabilities inadequately addressed. In algorithmic risk scoring, models may be tuned to achieve zero false negatives in a rare category at the cost of dramatically increasing false positives across more common categories, reducing overall system accuracy.
This page taught you the name. The deck turns the name into reflex. 1,100+ swipeable scenarios, 1,100+ defenses, 650+ detection prompts — spaced-repetition Swipe Deck, unlimited Spot-the-Bias Quiz, Defense Playbook, Pre-Flight, My Blindspots, Cheat Sheets, Field Guide e-book. $39.53$59.
