SandboxAQ Adds Trust and ROI Tools to Switch AI Agent Platform

SandboxAQ on October 19, 2026, announced an expansion of its Switch AI platform for enterprises adopting AI agents, introducing Switch Trust, for finding, testing, and guarding agents, and Switch ROI, which tracks the cost and value of each agent.

The Palo Alto, California-based company said the two products extend its open-source Switch Rooms offering, positioning them as a vendor-neutral way for organizations to assess agent risk, measure business value, and tune performance across whichever agents, models, and providers they use. The announcement contrasts agents with the first generation of enterprise AI, which it says was measured mainly through seats, usage, and token consumption; because agents can make decisions, take actions, and work alongside people, it says, they create new expectations for oversight and accountability.

Switch Rooms brings employees and AI agents together across Slack, Microsoft Teams, Discord, and other workplace tools, letting teams work with the agents they choose while keeping ownership of shared context, history, and knowledge. SandboxAQ launched Switch in August 2026 as a Commons Clause Apache 2.0 open-source offering, based on the premise that companies should be able to choose and change AI agents, models, and providers without giving up control of their context.

Switch Trust: Finding, Testing, and Guarding Agents

According to the announcement, Switch Trust provides agent security and governance across four steps: Discover, Scan, Evaluate, and Monitor. The Switch Trust product page states the product works for every agent, regardless of who built it.

The Discover step builds an inventory of the agents, models, tools, MCP servers, and dependencies running across connected environments and code repositories, including agents that security teams may not know exist. It records each item’s ownership, source, dependencies, findings, and severity to give teams a current view of their agent footprint.

The Scan step checks agents in code and in cloud environments for security vulnerabilities, combining static analysis with AI-assisted triage that filters out false positives. It assigns severity scores, maps supported findings to OWASP agentic security categories, and recommends fixes.

The Evaluate step connects to an agent directly and runs adversarial and functional probes against it, covering prompt injection, jailbreaks, data leakage, and instruction adherence. Each attack receives a pass or fail result that rolls into an overall health score, and teams can add their own evaluations alongside the built-in tests. The product page states that a failed test holds an agent’s release until it passes.

The Monitor step protects agents in production. Teams onboard an agent through a software development kit that records its activity, including prompts, responses, model calls, latency, sessions, and policy events, while agents running in Switch Rooms receive guardrail policies automatically. In live interactions, the policies look for prompt injection, secrets, personally identifiable information, toxicity, and jailbreaks, and Trust can block, redact, alert on, or log unsafe behavior while preserving evidence of what happened. The product page states that guardrails check what goes into and out of each agent and that Switch Trust keeps a single record of what each agent did and who approved it.

Switch ROI: Per-Agent Cost and Value Measurement

According to the announcement, Switch ROI measures the cost and the delivered value of every agent, paired with specific recommendations to lower cost and improve performance. The Switch ROI product page states it covers every agent from any vendor in one place, consolidating AI spend that is spread across seats, usage-based bills, and vendors.

The Measure feature follows spending and performance down to the individual agent, session, and turn, with budgets, burn rates, and alerts intended to surface overspending early. Rather than counting infrastructure metrics such as tokens or API calls, it ties that spending to outcomes including tickets resolved, pull requests accelerated, and tasks automated.

The Prove feature weighs the value each agent produces against its cost: Switch establishes a human baseline, tracks agent spending automatically, and measures figures such as cost per outcome and net return on investment.

The Optimize feature looks for leaking spend and decides what to fix, reroute, or trim, ranking recommendations for each agent with an estimated monthly savings range and a confidence level. The announcement lists the problem areas these recommendations address: poor model selection, tools that are offered but never used, context that is resent instead of cached, retries and loops that repeat work, and output that runs longer than the task needs. The product page states that the recommendations cover four kinds of cost improvement: model choice, unused tools, answer length, and caching. It says performance improvements are coming, and it describes moving an agent’s work to a cheaper model where quality checks pass while the agent keeps the room’s context.

Mount Sinai West and Executive Statements

According to the announcement, Mount Sinai West’s MSW Lab is using Switch Trust to build G Board, a framework that lets physicians create clinical applications with AI agents. “In the MSW Lab, physicians build real healthcare software with AI agents, so we have to prove those agents behave safely,” said Dr. Jeffrey Bander, Chief of Cardiology at Mount Sinai West and Associate Professor of Cardiology and Artificial Intelligence at the Icahn School of Medicine at Mount Sinai, in SandboxAQ’s announcement. Bander said Switch AI gives the lab a way to build that trust into everyday work, letting physicians concentrate on clinical problems instead of becoming AI security specialists.

The announcement cites a Gartner prediction that more than 40 percent of agentic AI projects will be canceled by the end of 2027. The Switch ROI product page separately cites KPMG Global AI Pulse survey data dated June 24, 2026, drawn from 2,145 respondents, reporting that 7 percent of leaders have an established return on AI and that leaders with clear AI cost visibility are five times as likely to report one.

Mohammed Aboul-Magd, general manager of Switch AI at SandboxAQ, said the company has seen cases across many users in which AI spending expected to avoid a net-new hire quickly became more expensive than hiring a full-time person. He said adopting AI is not by itself a return on investment; teams need to know each agent’s cost, what it produces, and how to improve it, which he described as the problem Switch ROI is built to solve.

Jack Hidary, SandboxAQ’s CEO, said companies will have hundreds, and potentially thousands, of AI agents working alongside their employees. He said companies would not hire a thousand people without knowing their qualifications, cost, or results, and that the same discipline needs to apply to agents; as the technology advances, he said, companies must also be able to move to better agents and models without losing the knowledge they have accumulated.