Safety Signalhttps://aisafetysecurity.org/The people, ideas and evidence shaping safe AI.The prompt filter is not your agent's security boundaryhttps://aisafetysecurity.org/articles/prompt-filters-are-not-your-agent-security-boundary/https://aisafetysecurity.org/articles/prompt-filters-are-not-your-agent-security-boundary/2026-09-27A CISO's warning about filter-first defenses becomes more useful when tested against agent attack data: protect the action path, not just the text entering the model.