Arrakis Security is building a Behavioral Governance platform for the AI agents already operating inside the enterprise, the ones IT approved and the ones nobody did. Security teams can usually confirm an agent exists or is connected to a system, but not what it is actually doing, which permissions it is using, or when ordinary behavior crosses into risk. Arrakis calls this the visibility illusion, and it's the gap the platform is built to close.
The platform works in three stages. It first discovers and inventories every agent across SaaS, cloud, and endpoint environments, building a behavioral baseline for each one. From there it moves to pre-execution controls: static analysis, MCP gateway enforcement, and policy allow-lists that catch risk before an agent acts. The third stage treats every agent output as untrusted, watching for anomalous intent and the kind of cross-agent contagion that can spread through a workflow before a human notices. Arrakis Labs, the company's research arm, is staffed by former Microsoft security researchers and veterans of top-tier technology units, feeding new detection patterns into the platform as agent-specific threats emerge.
Arrakis was founded by Tal Baron (CEO), Omer Efrat (CPO), and Ron Shani (CTO), veterans of Torq and Palantir who spent years building enterprise automation platforms and data infrastructure for large organizations. That vantage point showed them how far agent adoption had already outpaced the tools meant to secure it, and how little of the existing security stack was built to understand autonomous behavior rather than human clicks.
