Your staff paste pricing into public AI tools. Decide what data AI can touch, and under what conditions.
By Chris McCarthy, Founder & Chief AI Officer, OB.1
Your people are already using AI at work, including tools you never approved. The way to protect pricing, drawings and customer data is to answer one question in writing: what data can the AI access, and under what conditions? Then give people an approved tool that makes following the rule easier than breaking it.
This is part four of our Pilot to Production series. This one is for CIOs, IT directors and security leads.
Shadow AI is already in your building
An estimator pastes a customer's bid package into a public chatbot to summarize it. An engineer uploads a drawing to ask about a tolerance. A sales rep drops last year's price list into a tool to draft a quote letter. Each one is trying to do the job faster. Each one may have just sent your company's data somewhere you do not control.
This is common. In the 2024 Work Trend Index from Microsoft and LinkedIn, 75% of knowledge workers said they use AI at work, and 78% of AI users said they bring their own tools. In Cisco's 2024 Data Privacy Benchmark study, 48% of respondents admitted entering non-public company information into generative AI tools.
Banning AI rarely stops this. It moves the activity to personal phones and personal accounts, where you see even less.
What is actually at risk for a manufacturer
The exposure is specific. For most mid-sized manufacturers, the data that matters most falls into a few groups:
- Pricing, margins and quote history
- Engineering drawings, specs and process recipes
- Customer and supplier contracts
- Employee records
- Controlled technical data, for shops that do defense or export work
The cost is measurable. IBM's 2025 Cost of a Data Breach report found that one in five organizations reported a breach due to shadow AI. Organizations with high levels of shadow AI saw breach costs $670,000 higher on average. Of the organizations that reported a breach of an AI model or application, 97% said they had no AI access controls in place.
The question to ask: what data can the AI access, and under what conditions?
Answer it as a short table your whole company can read. Sort your data into tiers, then say which tools may touch each tier.
| Data tier | Examples | AI rule |
|---|---|---|
| Public | Website copy, published catalogs | Any approved tool |
| Internal | Procedures, meeting notes | Company-approved tools only |
| Confidential | Pricing, drawings, contracts | Approved tools with no training on your data and logged access |
| Restricted | Controlled technical data, personnel files | No AI use without written sign-off |
Then name the conditions. Which accounts are approved. Whether the vendor may train on your data. Where logs live and who reviews them. Who grants an exception, and how fast.
What a good answer looks like
A policy people follow has four parts:
- A one-page acceptable use policy in plain words, with real examples from your plant
- An approved tool that covers the common jobs, so people have a safe place to go
- Access controls and logging that match the data tiers
- A named owner who reviews exceptions and updates the rules as tools change
The NIST AI Risk Management Framework gives you a recognized structure for this work. Use it as a reference, sized to a mid-sized company.
How OB.1 fixes this
The Operational Blueprint starts by finding out how AI is already being used, without blame. We talk to the people doing the work and ask what they use and why. That shows you the real exposure and the jobs your approved tool needs to cover.
From there you get data tiers built around your actual information, access rules leadership can sign, and a plain-language acceptable use policy. Our Govern service then runs those controls: policy management, monitoring, escalation design and quarterly reports your leadership can cite. Our Rules Before Tools training gets your people using AI inside the lines.
Read the rest of the series
- Costs climb faster than your prices
- Your ERP data sits in silos nobody trusts
- The plant you just bought runs a different system
- Staff paste pricing into public AI tools (this article)
- Your best tech retires, and the fixes leave too
- Automation could ship the wrong order
- Teams lean on the system and lose the judgment to step in
Sources
- Microsoft and LinkedIn, 2024 Work Trend Index, May 2024
- Cisco, 2024 Data Privacy Benchmark Study, January 2024
- IBM, Cost of a Data Breach Report 2025, July 2025
- NIST, AI Risk Management Framework, AI RMF 1.0 (January 2023) and the Generative AI Profile, NIST AI 600-1 (July 2024)