Catastrophizing Bias

A system can treat a genuinely unlikely bad outcome as though it were nearly certain. That treatment makes small risks look much bigger than they are and drives an overreaction.

This bias overweights a low-probability adverse event, producing a disproportionate inferred severity. It systematically overestimates negative outcomes during evaluative aggregation and decision weighting.

A news app's recommendation engine notices you clicked on one story about a rare disease outbreak. It then floods your feed with dozens of alarming health articles, treating that single click as proof you're highly interested in worst-case scenarios — even though the actual risk of any outbreak affecting you is extremely small.

A content ranking pipeline applies a heavy multiplier to items flagged with adverse-outcome features like injury or financial loss. An item with a true base-rate probability of 0.4% adverse relevance ends up scored as though it were more like 1.4%, crossing the threshold for promoted placement. Over successive ranking cycles, that catastrophe-framed content accumulates clicks and keeps reinforcing its own elevated score. An audit later reveals the top 10% of promoted slots are occupied by adverse-outcome content at more than four times its actual population rate — a direct consequence of catastrophizing baked into the ranking logic.

Any hint of danger gets boosted far more than an ordinary signal would. That boost is exactly what makes a rare danger get treated like a common one.

Negative-oriented features receive higher weight than other candidate features, producing an asymmetric ranking across items. That imbalance lets low-probability adverse signals dominate the selection and scoring process.

Lowering the extra weight given to negative signals is the direct fix. Checking the actual odds before reacting keeps the reaction proportional to the real risk.

Calibrating the negative-signal weight against empirical base rates corrects the overamplification directly. Probabilistic calibration and threshold adjustment align the visibility with the true likelihood involved.

Overprioritizing unlikely negatives; False alarm proliferation; Resource drain on mitigation

An adversarial actor can deliberately seed low-probability negative signals into content metadata or engagement patterns to trigger catastrophizing bias in a platform's visibility allocation pipeline, causing benign content to be suppressed while threat-framed narratives receive outsized distribution. By exploiting the asymmetric gain applied to adverse-event features, a bad actor can manufacture salience for fabricated risks—amplifying panic-inducing content that out-competes factual material in ranking and slot allocation. This mechanism can also be used to repeatedly trigger false-alarm proliferation, draining moderator and algorithmic resources away from genuine harms and toward manufactured ones.

Introduce empirical base-rate calibration layers into the visibility_filtering module so that negative-signal gain is regularized against historical frequency data rather than raw feature salience. Conduct periodic cohort-stratified exposure audits to identify systematic overrepresentation of low-probability adverse content in top-ranked slots, and apply decay-constant recalibration to normalize adverse-signal weighting over time. Training human reviewers and algorithmic auditors to surface feature-weight asymmetry in scoring logs provides an additional checkpoint against runaway negative amplification.