← Back to Blog

Microsoft Copilot in Manufacturing: When AI Meets OT Data

E2E Agentic Bridge·March 4, 2026

Manufacturing Has a Copilot Problem Nobody's Talking About

The manufacturing sector is adopting Microsoft 365 Copilot faster than most industries. And why not? Summarizing supplier contracts, drafting quality reports, analyzing production data in Excel — the productivity gains are real. A 2025 study published in the Proceedings of the Annual Conference of the Prognostics and Health Management Society showed that LLM pipelines can already extract problems and solutions from ERP machine history, cluster downtime causes, and accelerate root-cause analysis.

But manufacturing environments have something most office-centric industries don't: operational technology (OT) data that crosses into IT systems. Production schedules, SCADA reports, PLC configurations, supplier pricing, and quality control metrics increasingly live in SharePoint, Teams, and Excel — exactly where Copilot looks for answers.

When a plant engineer asks Copilot to "summarize this week's production issues," the response might include data from maintenance logs, supplier communications, quality holds, and equipment configurations. That's useful for the engineer. It's also a data exposure risk that most manufacturing IT teams haven't accounted for.

The IT/OT Convergence Problem

Manufacturing environments traditionally separated IT (business systems) from OT (industrial control systems). The air gap between the corporate network and the plant floor was both a security boundary and a data classification boundary.

That air gap is gone. Modern manufacturing operations push data from OT systems into M365 through:

  • Power Automate flows that pull SCADA alerts into Teams channels
  • SharePoint document libraries storing equipment manuals, PLC configurations, and maintenance procedures
  • Excel workbooks with production data exported from MES (Manufacturing Execution Systems)
  • Teams channels where plant floor supervisors discuss equipment issues, share photos, and coordinate shifts
  • Power BI dashboards connected to historian databases (which Copilot in Power BI can query)

Every one of these data flows is now within Copilot's reach. And unlike traditional IT data, OT data exposure can have physical consequences.

What's Actually at Risk

Production Schedules and Capacity Data

Production schedules reveal what you're making, how much, and when. For manufacturers in competitive industries — automotive, electronics, pharmaceuticals — this is commercially sensitive intelligence. A competitor knowing your production capacity, shift schedules, and order volumes could undercut pricing or time product launches.

Copilot can surface production schedule data from any Excel file, SharePoint list, or Teams conversation that mentions it. If a plant manager shares next quarter's production plan in a Teams channel, Copilot can summarize it for anyone with access to that channel — including contractors, temporary workers, or employees who changed roles but retained old Teams memberships.

Supplier Contracts and Pricing

Manufacturing supply chains run on negotiated pricing. Your contract with a steel supplier, a component manufacturer, or a logistics provider contains pricing that both parties consider confidential. These contracts typically live in SharePoint document libraries.

Copilot can read, summarize, and compare these contracts. An employee asking "what are our current shipping rates with FedEx?" could get an answer that includes negotiated rates from a contract they weren't involved in negotiating. Worse, if a supplier's employee has guest access to your tenant (common in collaborative manufacturing relationships), they could potentially use Copilot to discover pricing from their competitors' contracts stored in your tenant.

Equipment Configurations and Maintenance Data

PLC configurations, SCADA parameters, and equipment maintenance procedures are increasingly digitized and stored in M365. This data is sensitive for two reasons:

  1. Competitive advantage — custom equipment configurations and optimized process parameters represent years of engineering investment
  2. Safety and security — detailed knowledge of industrial control system configurations is exactly what threat actors need to plan attacks on OT infrastructure

The 2025 EchoLeak vulnerability (CVE-2025-32711) demonstrated that Copilot could be exploited to exfiltrate data through prompt injection. In a manufacturing context, an attacker injecting prompts through a shared document could potentially extract SCADA configurations or PLC parameters that were stored in adjacent SharePoint libraries.

Quality Control and Compliance Records

Quality holds, non-conformance reports, customer complaints, and FDA/regulatory submissions contain information that's both commercially sensitive and legally protected. A Copilot response that surfaces a quality hold from three months ago — one that was resolved but never removed from SharePoint — could create legal liability if it reaches the wrong audience.

For pharmaceutical manufacturers under FDA regulation, or food manufacturers under FSMA, quality records have specific access control requirements. Copilot doesn't understand regulatory access requirements — it only understands SharePoint permissions.

Manufacturing-Specific Copilot Controls

Segment Your SharePoint Sites

The first and most effective control: don't store OT-related data in the same SharePoint sites as general business content.

Create dedicated SharePoint sites for:

  • Plant Operations — production schedules, shift reports, equipment logs
  • Engineering — PLC configurations, SCADA documentation, process parameters
  • Quality — NCRs, quality holds, audit reports, regulatory submissions
  • Supply Chain — supplier contracts, pricing agreements, logistics data

Apply strict permissions to each site. The marketing team doesn't need access to engineering documentation. The HR team doesn't need to see supplier contracts. And Copilot can only surface content the user has permission to access.

This sounds basic because it is. But a 2025 Microsoft report found that the average M365 tenant has over 40,000 SharePoint sites, and most manufacturing organizations haven't reviewed permissions since initial deployment. Run a SharePoint permissions audit before Copilot goes live.

Apply Sensitivity Labels to OT Data

Sensitivity labels are your second line of defense. Even if permissions are misconfigured, sensitivity labels with associated DLP policies can prevent Copilot from surfacing protected content.

For manufacturing, create sensitivity labels specific to your data types:

  • OT-Confidential — SCADA configurations, PLC programs, control system documentation
  • Commercial-Confidential — supplier pricing, production volumes, capacity data
  • Quality-Restricted — NCRs, regulatory submissions, audit findings
  • Engineering-Internal — process parameters, equipment specifications, maintenance procedures

Configure DLP policies so that content labeled OT-Confidential or above is excluded from Copilot responses. Users can still access the files directly, but Copilot won't summarize, reference, or include them in AI-generated responses.

Lock Down Teams Channels

Plant floor Teams channels are goldmines of operational data. Supervisors share photos of equipment issues. Maintenance techs discuss PLC configurations. Quality managers flag production holds. All of this becomes Copilot-searchable.

For sensitive Teams channels:

  1. Use private channels for discussions involving OT data
  2. Apply sensitivity labels to channels containing operational information
  3. Regularly audit channel membership — remove contractors and former employees
  4. Consider information barriers if your organization has teams that shouldn't see each other's operational data (e.g., competing product lines)

The Teams channel cleanup is often the highest-impact, lowest-effort action for manufacturing Copilot security. Our guide on M365 Group cleanup covers the process in detail.

Control Copilot in Power BI

Many manufacturers connect Power BI to operational databases — historian systems, MES databases, ERP systems. Copilot in Power BI can query these datasets using natural language, which means an employee could ask "show me all equipment failures in the past month" and get detailed OT data they may not need.

Control this through:

  • Row-level security (RLS) in Power BI — restrict which data rows users can see based on their role
  • Object-level security (OLS) — hide sensitive columns (equipment serial numbers, specific process parameters)
  • Workspace permissions — limit who can access Power BI workspaces containing OT data
  • Copilot feature toggles — Power BI admin settings allow disabling Copilot for specific workspaces

Guest Access Restrictions

Manufacturing supply chains involve extensive collaboration with suppliers, customers, logistics providers, and contract manufacturers. Many of these partners have guest access to your M365 tenant.

Guest users with Copilot access can potentially discover information far beyond what they were explicitly shared. A supplier with guest access to a project-specific Teams channel might use Copilot to surface documents from other areas of your tenant where permissions are overly broad.

Our guest access risks guide covers the broader problem. For manufacturing specifically:

  • Review all guest user permissions — especially in SharePoint sites containing operational data
  • Apply Conditional Access policies requiring managed devices for guest Copilot access
  • Use information barriers to prevent guest users from discovering content outside their designated collaboration areas
  • Set expiration policies for guest accounts — manufacturing project timelines are finite, but guest access often isn't

The Physical Safety Dimension

This is where manufacturing Copilot risks diverge from every other industry. In a financial services firm, a data breach costs money and reputation. In a manufacturing environment, exposed OT data can create physical safety risks.

Consider: an attacker who obtains SCADA configurations and PLC parameters from a Copilot-assisted data exfiltration could:

  • Identify vulnerabilities in industrial control systems
  • Understand safety interlocks and their bypass conditions
  • Map the relationship between IT and OT networks
  • Plan targeted attacks on specific equipment or processes

This isn't theoretical. The ICS-CERT has documented increasing targeting of manufacturing OT systems, and the convergence of IT and OT data in M365 creates a new attack vector that most manufacturing security teams haven't modeled.

Regulatory Considerations

Manufacturing compliance requirements vary by sub-industry:

  • Pharmaceutical (FDA 21 CFR Part 11) — electronic records must have access controls, audit trails, and electronic signatures. Copilot accessing these records without proper controls may violate Part 11 requirements.
  • Food (FSMA) — food safety records have specific access and retention requirements
  • Automotive (IATF 16949) — quality management records must be controlled and traceable
  • Aerospace (AS9100) — configuration management and document control requirements apply to any AI accessing controlled documents
  • Chemical (EPA/OSHA PSM) — process safety management data has strict access controls

None of these regulations were written with AI access in mind. But regulators are starting to ask questions. If your quality records are accessible through Copilot without appropriate controls, your next audit may not go as smoothly as the last one.

Implementation Roadmap for Manufacturing

Phase 1 (Weeks 1-2): Discovery

  • Inventory all OT-related data in M365 (SharePoint, Teams, OneDrive, Power BI)
  • Map data flows from OT systems into M365
  • Identify all guest users with access to operational data
  • Document which Teams channels contain plant floor discussions

Phase 2 (Weeks 3-4): Classification and Permissions

  • Create manufacturing-specific sensitivity labels
  • Apply labels to OT data repositories
  • Clean up SharePoint permissions — remove overly broad access
  • Audit and prune Teams channel memberships

Phase 3 (Weeks 5-6): Policy Deployment

  • Deploy DLP policies preventing Copilot from surfacing OT-labeled content
  • Configure Conditional Access policies for Copilot (device compliance, location restrictions)
  • Set up Power BI row-level and object-level security
  • Restrict guest user Copilot access

Phase 4 (Ongoing): Monitoring

  • Enable Purview audit logging for Copilot interactions
  • Monitor for Copilot accessing OT-related content
  • Review and update policies quarterly as Copilot capabilities evolve
  • Train plant floor staff on what not to share in Copilot-accessible locations

The Bottom Line

Manufacturing organizations can't treat Copilot deployment as a standard IT rollout. The presence of OT data in M365 — production schedules, equipment configurations, quality records, supplier pricing — creates risks that don't exist in purely office-centric environments.

The controls aren't complicated: proper permissions, sensitivity labels, DLP policies, and Conditional Access. But they need to be applied with an understanding of manufacturing data flows that most generic Copilot deployment guides don't cover.

Your plant floor data is now in the same system as your AI assistant. Make sure they're properly introduced.

Take Action Now

Manufacturing firms face unique Copilot risks that generic assessments miss. Run a free scan to identify OT data exposure in your M365 tenant and get manufacturing-specific remediation priorities.