Switch AI adds free tools to inventory agents, block injections, and track ROI

SandboxAQ has expanded its Switch AI platform with two new products designed to improve security and financial oversight of artificial intelligence agents. The company announced the launch of Switch Trust and Switch ROI on Monday, aiming to help enterprises manage the risks and costs associated with deploying autonomous AI systems.

What happened

SandboxAQ added Switch Trust and Switch ROI to its existing Switch AI platform, which originally launched in August. Switch Trust focuses on security by first identifying all AI agents, models, and tools running within a company’s environment, including those unknown to security teams. It then scans these components for vulnerabilities and tests them against specific attacks, such as prompt injection, where hidden instructions cause an agent to behave incorrectly.

Once an agent is operational, Switch Trust monitors input and output traffic. The tool can block or mask sensitive information, such as leaked passwords or personal data, and maintains a log of every action taken by each agent. This provides visibility into agent behavior that might otherwise go unnoticed by internal security staff.

Switch ROI addresses the financial side of AI deployment by tracking the cost of each agent and correlating that spend with tangible results. These results include metrics like the number of support tickets closed or tasks automated. The tool also recommends ways to reduce expenses, such as shifting workloads to cheaper AI models or removing unused tools from an agent’s workflow.

Why it matters

The launch addresses growing concerns about both the security and value of AI agents. SandboxAQ noted that AI spending intended to replace a new hire often exceeds the cost of a full-time employee. Mount Sinai West’s MSW Lab is currently using Switch Trust while its doctors develop medical software with AI agents, according to the company.

Broader industry data highlights the difficulty in realizing value from agents. Gartner predicted that over 40% of agentic AI projects will be cancelled by the end of 2027. Similarly, AWS stated that nearly 90% of Amazon’s early agent prototypes never shipped. Security risks are also rising, with companies like Google and JPMorgan recently fixing flaws in the Model Context Protocol (MCP), a standard for linking agents to tools.

What to watch

• Monitor for updates on pricing tiers beyond the free five-agent limit.
• Watch for new prompt injection techniques that may bypass current testing.
• Track industry adoption rates of MCP and related security standards.
• Observe how healthcare and other regulated sectors implement agent oversight.

Who needs to act first

Teams should immediately inventory all AI agents currently operating in their environment, including those deployed by individual developers without central oversight. Security leaders need to verify whether these agents are exposed to prompt injection risks and if sensitive data is being transmitted unencrypted. Organizations should establish baseline metrics for agent performance and cost before deploying new tools. If agents are already live, teams must ensure that logging is enabled to capture input and output for audit purposes. Prioritize reviewing agents that handle personal data or financial transactions, as these present the highest risk for data leakage and unauthorized actions.

What to do and how to stay safe: SandboxAQ

• Conduct a full inventory of all AI agents, models, and tools currently in use.
• Test existing agents for prompt injection vulnerabilities using available tools.
• Implement monitoring to block or mask sensitive data in agent inputs and outputs.
• Track agent spending against specific business outcomes to identify inefficiencies.

Step-by-step guide: Virtual Patching Myths: What It Actually Fixes and Breaks

General security guidance from the Patch Gazette newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

What is Switch Trust?

Switch Trust is a tool that inventories AI agents, tests them for security flaws like prompt injection, and blocks unsafe content such as leaked passwords.

Is Switch ROI free?

Both Switch Trust and Switch ROI are free for up to five managed agents, according to a company blog post that did not list paid prices.

Who is using Switch Trust?

Mount Sinai West’s MSW Lab is using Switch Trust while its doctors build medical software with AI agents, SandboxAQ said.