Canyon Road

Canyon Road provides execution-layer security for AI workloads, focusing on runtime control over what AI systems can access, execute, and connect to. The company's approach targets two distinct execution contexts: supervised AI running on endpoints and unsupervised agents operating in automated environments. The product portfolio consists of three components: - Beacon: A security agent for supervised AI on endpoints (e.g., Cursor, Claude Desktop, ChatGPT). It adds guardrails and approval workflows to AI copilots and desktop tools running under employee credentials, keeping endpoints both productive and compliant with policy. - AgentSH: An open-source tool that enforces least-privilege security at the syscall level for headless, unsupervised agents. It is designed for use in CI pipelines, containers, and development environments where no human is present to supervise AI actions. - Watchtower: A centralized command plane that distributes policy to Beacon and AgentSH, routes approval requests, exports audit logs to SIEM systems, and provides a fleet-wide kill switch. Canyon Road's enforcement model allows, prompts, blocks, or redirects AI-triggered connections, commands, file changes, and tool calls based on configurable policy. A key design principle is "steering" — redirecting AI activity toward approved registries or endpoints rather than issuing hard blocks, which can cause retry loops in agent workflows. The target market includes organizations where developers have deployed AI coding tools and where AI agents are being integrated into internal automation. The company positions its solution against the visibility and control gaps that arise when AI executes actions inside a corporate environment without IT provisioning or security oversight.