●builderYou should implement these red-teaming patterns to secure agentic tools that possess real-world permissions.
●policyThese findings highlight the urgent need for safety standards specifically targeting multi-agent autonomy and tool-use vulnerabilities.