Menlo Security: Real-time AI Agent Security & Prompt Injection

Menlo Security is tackling real-time AI agent security with its MARS platform. This comes as AI agents are increasingly accessing enterprise systems and sensitive data. The company aims to provide greater visibility into agent actions and enforce policies in real time. Ramin Farassat, chief product officer of Menlo Security, highlights the risks. He explains that agents can be "fooled with commands and prompts through prompt poisoning" to perform unintended tasks. This is because AI agents often don't distinguish between legitimate data and malicious commands. Menlo Security's approach involves setting policies and controlling agent actions. They also work to remove sensitive information from the data agents consume, masking it in real time. Farassat notes that no single prompt injection technology will be 100% safe, emphasizing it's one tool in a larger security strategy. This matters because protecting sensitive data from AI agent vulnerabilities is crucial for businesses today.