SandboxAQ launches Switch Trust and Switch ROI to secure AI agents

SandboxAQ launched two new products on Monday to help organizations secure their artificial intelligence agents and measure their financial return. The company introduced Switch Trust and Switch ROI as additions to its existing Switch AI platform, which originally debuted in August. These tools aim to address growing concerns about agent security and cost efficiency.

Switch Trust focuses on security by first identifying all AI agents, models, and tools operating within a company’s systems. This includes discovering assets that security teams may not be aware of. The tool then scans these components for vulnerabilities and tests them against specific attack vectors, such as prompt injection, where hidden instructions manipulate agent behavior.

Once an agent is active, Switch Trust monitors inputs and outputs to detect unsafe content. It can block or mask leaked passwords, personal data, and other sensitive information. The system also maintains a detailed record of each agent’s actions, providing an audit trail for compliance and security review purposes.

Financial tracking and vendor neutrality

Switch ROI addresses the economic side of AI deployment by tracking the cost of each individual agent. It links this spending directly to measurable business results, such as the number of support tickets closed or tasks automated. The tool provides recommendations for reducing expenses, including shifting workloads to cheaper models or removing unused tools.

Both products are vendor-agnostic, meaning they work with AI agents from any provider. According to a company blog post, both tools are free for organizations managing up to five agents. The post did not disclose pricing tiers for larger deployments or additional managed agents beyond this initial limit.

Industry context and adoption

Mohammed Aboul-Magd, who leads the Switch AI division, noted that companies often find AI spending intended to replace hires exceeds the cost of a full-time employee. To demonstrate real-world application, SandboxAQ stated that Mount Sinai West’s MSW Lab is using Switch Trust while its doctors develop medical software with AI agents.

The launch coincides with broader industry struggles to derive value from agentic AI. Gartner predicted that more than 40% of agentic AI projects will be cancelled by the end of 2027. Additionally, AWS reported that nearly 90% of Amazon’s early agent prototypes never reached production.

Security risks are also escalating across the sector. Major organizations including Google and JPMorgan have recently patched the same vulnerability in the Model Context Protocol, a standard used for connecting agents to external tools. This highlights the increasing need for robust security testing in AI deployments.

What should you do?

Organizations deploying AI agents should conduct a complete inventory of all models and tools in their environment. Security teams should test these agents for prompt injection vulnerabilities before allowing them to process sensitive data. Companies should also establish clear metrics to track agent costs against specific business outcomes to ensure financial viability.

Background: Prompt injection attacks

Look for sudden shifts in tone, refusal to follow system constraints, or output that repeats hidden instructions. These signs indicate the model is processing injected commands rather than your intended query.

Read the full guide: Prompt Injection Warning Signs: How to Spot AI Manipulation

What to do and how to stay safe: SandboxAQ

• Inventory all AI agents and models currently running in your environment to identify unknown assets.
• Test agents for prompt injection vulnerabilities and other security flaws before deploying them to production.
• Monitor agent inputs and outputs to prevent the leakage of passwords, personal data, and other sensitive information.
• Track the specific costs of each agent and link them to measurable business results like tasks automated or tickets closed.

Step-by-step guide: Risk-Based Vulnerability Management: Prioritize Fixes by Actual Danger

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

Frequently asked questions

What is the cost for using Switch Trust and Switch ROI?

Both products are free for organizations managing up to five agents, according to a company blog post.

How does Switch Trust detect unknown AI agents?

It lists all AI agents, models, and tools running in a company’s systems, including those the security team may not know about.

What security vulnerability does Switch Trust test for?

It tests agents against attacks such as prompt injection, where hidden instructions trick an agent into misbehaving.