Data hazards occur in pipelined processors when instructions that exhibit data dependencies are executed in parallel, leading to incorrect results or delays. Effective handling of data hazards is crucial for maintaining the efficiency and correctness of instruction execution in modern CPU architectures.