SandboxAQ Adds Agent Security and ROI Tools to Switch AI

SandboxAQ expanded Switch AI with Switch Trust and Switch ROI, the company announced in a press release. The tools help businesses assess AI agent security, monitor behavior, and compare costs with business outcomes across models and providers.

Switch Trust inventories agents, models, tools, MCP servers, and dependencies across connected environments and code repositories. It scans for vulnerabilities, tests agents with prompt injection and jailbreak attempts, and monitors prompts, responses, model calls, latency, sessions, and policy events. Its controls can block, redact, flag, or record activity involving secrets, personal information, toxic content, and other policy violations.

Switch ROI tracks spending and performance by agent, session, and interaction. It connects costs to outcomes such as resolved tickets, faster pull requests, and automated tasks, then calculates cost per outcome and net return. The tool also identifies potential savings from model selection, unused tools, repeated context, retries, loops, and excessive output.

Both products extend Switch Rooms, an open source system that lets people and agents share company controlled context through Slack, Microsoft Teams, Discord, and other workplace tools. Mount Sinai West's MSW Lab is using Switch Trust to build G Board, a framework for physicians creating clinical applications with AI agents.