Paper compares budgeted scheduling for LLM jailbreak red-teaming
Historical success priors, semantic guidance, and online feedback were tested across three models and three defenses on GPT-5.2.
Stories per day · 30 days
1 stories, newest first · one card per story