This research introduces the jailbreak tax - the quality degradation that occurs when LLMs are forced to bypass safety guardrails.

• Jailbreaks that bypass model safety measures produce significantly lower quality outputs
• Quality degradation (the 'tax') averages 13-33% across tested models
• Even when models appear successfully jailbroken, the utility of their harmful outputs is questionable
• Evaluation conducted using specially designed benchmarks with verifiable ground truth

For security professionals, this research suggests inherent safeguards in LLMs beyond explicit guardrails, as models struggle to produce high-quality harmful content even when safety systems are bypassed.

The Jailbreak Tax: How Useful are Your Jailbreak Outputs?