A negative value indicates that the item may be damaging the overall psychometric reliability of the measure.
[3][4] Identifying and removing (or revising) poorly-performing items is a critical way that psychometric analysis can improve the quality of a measure.
This is considered important because items vary in difficulty and the point-biserial correlation cannot attain its theoretical maxima [+1,-1] unless the proportion correct is 0.50 (50% answering the item correctly).
[5] The item-reliability index (IRI) is defined as the product of the point-biserial item-total correlation and the item standard deviation.
In practice, a negative IRI indicates the relative degree which an item damages the reliability estimate and a positive value indicates the relative degree which it contributes towards a high reliability estimate.