Somansa Extended Endpoint DLP to AI Agents, Enforcing File Access at the OS Permission Layer Rather Than the Network
Korean DLP vendor Somansa launched Privacy-i AIDR on August 26, a module inside its existing endpoint agent that inventories every AI agent on a machine, including locally-run Hugging Face models under Ollama or LM Studio that generate no cloud traffic, and then enforces execution permissions at the OS access-control layer. An admin can grant a specific agent read access to a project folder while denying the HR directory even though the human user account can reach both, and can block subprocess spawning or off-list network destinations. Somansa's research head puts the company roughly three months behind better-funded Western AIDR vendors, and the structural point is that endpoint DLP already owns the enforcement pipeline and only needs agent identity attribution bolted on.
Source
↳ Follow the thread