See and Stop Agent Risk at Runtime

OnDuty's AI Detection & Response (AIDR) capabilities are purpose-built to monitor agent intent and execution paths at runtime. By observing how agents reason, act, and chain decisions in real environments, AIDR enables security teams to detect and stop risky behavior before it impacts systems.

Threats detected

• Prompt Injection — Detects direct and indirect prompt injection that causes agents to execute unsafe actions or alter control flow.
• Data Exfiltration — Real-time monitoring and blocking of sensitive data leakage through agent conversations, tool calls, and memory persistence.
• Privilege Escalation — Continuous monitoring for unauthorized permission increases and lateral movement between agents and systems.
• Multi-Agent Attacks — Threat detection across complex agent workflows to identify coordinated attacks and sophisticated threat patterns.
• Memory Poisoning — Real-time detection and prevention of malicious context injection and agent memory manipulation attacks.
• Tool Misuse — Automated blocking of unauthorized tool invocations and API calls that violate security policies or business rules.