Sentrix provides an end‑to‑end security platform that monitors and controls autonomous AI agents using coordinated “police” swarms. The system deploys patrols and investigators to detect and mitigate threats such as data leaks, prompt injection attacks, and backdoors, offering multi‑level escalation to keep AI behavior in line.
Funding
Funding not disclosed
Founders
Product
Problem
Autonomous AI agents can act unpredictably, leading to data leaks, prompt injection attacks, and hidden backdoors that compromise security and privacy. As AI capabilities grow, detecting and containing rogue behavior in real time becomes increasingly difficult for organizations.
Solution
Sentrix offers an end‑to‑end security platform that deploys “agentic police” swarms to monitor and police autonomous AI systems. The platform mimics real police investigation teams, using multiple escalation levels to identify suspicious actions, isolate compromised agents, and remediate threats without interrupting normal operations. Its investigative agents are built on Nvidia’s Nemotron 3.3 30B model, fine‑tuned with synthetic data generation pipelines to recognize and respond to security violations within milliseconds. By integrating directly with existing AI deployments, Sentrix provides continuous, automated oversight that prevents data exfiltration, prompt injection, and backdoor exploitation.
Target Audience
Primary customers are enterprises and platform providers that operate large fleets of autonomous AI agents, such as conversational assistants, autonomous bots, and generative AI services, and need robust security oversight.
Features
- Hierarchical “patrol and investigator” swarm architecture that escalates incidents from detection to containment and remediation
- Real‑time threat detection with sub‑100 ms response times for rogue agent behavior
- Nemotron‑based investigative agents fine‑tuned via LoRA on synthetic data, achieving high token‑level accuracy for security pattern recognition
- Automated synthetic data generation pipelines for continuous model improvement and adaptation to new attack vectors
- Simple command‑line integration to deploy the police swarm alongside existing AI agents
- Continuous monitoring and logging of agent actions for auditability and compliance reporting