Security Incidents

Coordinated prompt injection attack attempted to extract customer PII through the support chatbot

Agent attempted to call unauthorized file system tool to access documents outside scope

Agent memory contamination allowed data from HR agent to leak to customer support agent

Data analyst agent exposed customer email addresses in query results without proper masking

Statistical analysis revealed potential gender bias in candidate ranking scores

Attacker bypassed rate limits using multiple IP addresses to abuse customer-support-copilot

User attempted to extract system prompts by exploiting legacy debug commands

Red team successfully jailbroke staging agent using novel multi-turn technique