Protective Capacity Hallucination: When Large Language Models Claim Nonexistent Capabilities