4.7Measuring Jailbreak Resistance

Metric	Target	Measurement
Overall resistance rate	> 95%	% of jailbreak attempts properly refused
Resistance by category	> 90% per category	No single category below 90%
New payload resistance	> 80%	% of newly discovered payloads blocked on first test
Multi-turn resistance	> 85%	% of multi-turn escalation attempts blocked
Regression rate	0%	Previously blocked payloads that now succeed

Jailbreak testing is an ongoing practice, not a one-time assessment. New bypass techniques emerge weekly, and your test suite must evolve to match.