2 ms·Guardrails for offensive AI agents and why instructions are never enough3 points by snorbleck 3mo ago