St. Louis · Kansas City
AI Data Security by Design: Read-Only by Default, and Nothing Changes Without Your Approval
Connecting AI to your business data is safe when the AI can only do what you allow. ThrottleIQ's MCP architecture gives AI tools read-only access to one database in your own Azure tenant by default. Each query is logged with who asked and when, and your existing permissions still decide what each person sees.

Zero-Write AI
That is AI data security enforced at the API layer instead of by policy. We call the default Zero-Write AI: the AI has no insert, update or delete endpoints, so it cannot modify your data. If you want it to take an action, such as updating a record, we add that one action after you approve it, scoped to that task and logged.
The stack
Secure AI Architecture in Four Layers, Read From the Bottom Up
Your systems sit at the base, your people at the top, and a bridge that is read-only by default runs between them.
Layer 4: Users
Plain-English answers in the AI tool you already use, BI dashboards and automation triggers.
Layer 3: MCP server, read-only by default
It exposes queries that read data and, by default, nothing that writes it. Actions you approve are added one at a time, and every call is recorded.
Layer 2: Sync Engine
A normalised data store, encrypted at rest, updated on a schedule and when records change. It only ever reads from your systems.
Layer 1: Source systems
Financial software, Microsoft 365, Azure and Entra ID.
A single query
What Happens When Someone Asks a Question, From Query to Recorded Answer
Query in
A user asks a question in plain English. The AI turns it into a read query and sends it through the MCP layer. By default, a read query is the only kind it can send.
Read only
The MCP server runs the query against the normalised data store. In Zero-Write mode nothing it exposes can write, so nothing is written.
Answer out
The result returns to the user, recorded with a timestamp and the identity of whoever asked. Your data is never used to train the model.
Zero-Write AI
Read-Only AI, Enforced in the Design Instead of a Policy Document
Read-only AI means the AI can look at your data and has no way to change it. A policy says a tool shouldn't change a record. A design without write endpoints means it can't, so there is nothing for a badly worded prompt, a mistake or a misused account to call.
Read-only is the default, not a ceiling. If you want the AI to take an action, such as updating a record, we add that one action after you approve it, scoped to that task and logged. Everything you haven't approved stays impossible, not merely prohibited.
That difference is what an auditor, an insurer or a nervous board member wants to hear about.
Why the distinction behind Zero-Write AI matters →Your tenant
AI Data Security That Stays Inside Your Own Azure Tenant
The whole platform runs in your Microsoft environment. You own the database, the data and the keys, and ThrottleNet builds and manages it while keeping nothing on our side.
Because it is Microsoft-native, the controls you already rely on keep working: MFA, conditional access and DLP policies still apply, and people see only what they already have rights to. Every query, and every action you approve, is logged, attributable and exportable, so an auditor's question about who asked what has an answer.
Why your AI should run in your own tenant →People and policy
Where Read-Only AI Fits With AI Governance and Shadow AI
The architecture covers the platform. People are the other half. When staff paste client data into personal AI accounts, that data leaves your environment and nobody can see what was shared. Giving your team the same conversational AI inside the controls above is the practical answer to shadow AI.
The rest is permissions and policy. Copilot inherits each user's permissions, so a file shared too widely years ago is one question away. AI governance covers who uses which tools, for what, and with what access.
Read our guide to shadow AI → How we handle AI governance →Ask Us the Hard Security Questions Before Anything Connects
Bring your IT lead, your auditor's checklist or your own list of worries. A free feasibility study works through how the platform would sit on your systems and what it is worth to your business, with no obligation and no sales pressure.
No sales pressure. Yours to keep.
FAQ
AI Data Security: Questions to Ask Before You Connect Anything
Can the AI Change or Delete Our Data?
Not by default. In Zero-Write mode the MCP layer exposes read queries only, with no insert, update or delete endpoints, so the AI has no way to modify a record. If you want it to take a specific action, you approve that action, it is added for that task only, and every use is logged. Nothing else can be written.
What Is MCP?
MCP stands for Model Context Protocol, a standard way for AI tools to connect to data sources. In ThrottleIQ, the MCP server is the bridge between your AI assistant and your Intelligence Database: authenticated, auditable and read-only by default, so Copilot, Claude, ChatGPT or another assistant can query your data through one controlled door.
Who Can See the Query Log?
You can. Each query, and each action you have approved, is recorded with a timestamp and the identity of the person who asked, and the log is exportable, so your IT lead, compliance team or auditor can review who asked what and when. The log sits in your environment, under your control.
Does the AI Train on Our Data?
No. Your data is not used to train any AI model. The model reads from your Intelligence Database to answer a question and returns the result to the person who asked, and your business information stays in your environment, under your keys, whichever assistant your team uses to ask the question.
