Multicollinearity
High correlation between input features, which lets a regression's individual coefficients swing wildly across resamples while its actual predictions stay stable.
When two features are nearly collinear, many different coefficient splits between them produce almost the same combined prediction, so the exact split the optimizer lands on becomes highly sensitive to small changes in the data. The variance inflation factor quantifies this directly: regress a feature on the others, and VIF grows without bound as that feature becomes more predictable from the rest.
It matters for interpreting individual coefficients and matters far less for pure predictive accuracy, since the combined effect stays stable even when the individual pieces don't.