A robotics integrator commissioning a production line needs remote access to the manufacturing execution system, so an account gets created with broad rights and a password that does not expire. Unlike the commissioning project, the account does not have an end date, and it keeps authenticating long after the integrator's last site visit.
Accounts like that are the infrastructure AI runs on, and they are easy to overlook while adoption accelerates. Every AI assistant or machine learning model requires access to enterprise data, production systems, engineering repositories and operational technology. Each new connection creates identities, permissions and machine credentials that someone has to govern.
The Netwrix 2026 Data & Identity Security Report, based on responses from IT and security leaders worldwide, puts a figure on the risk that creates. Companies where AI has significantly increased the number of identities in their environment experienced a 43 percent breach rate during the past year, compared to 11 percent among organizations where AI had not materially expanded their identity footprint.
The challenge is sharper in manufacturing, where a single site can run business applications, engineering systems, and production equipment on one converged network. The findings below describe a sector adopting AI faster than it can account for the identities AI creates, and they point to a specific order of work.
Manufacturing's Visibility Gap Is Large
Across all industries, the Netwrix report found that 74 percent of organizations lack a unified view of sensitive data and the identities that can access it.
Manufacturing organizations perform worse on several of these foundational capabilities. Only 34 percent report having a complete inventory of where sensitive data resides across their environment, compared to 43 percent across all industries, and only 21 percent can immediately determine which identities have access to sensitive data, compared to 30 percent overall.
AI systems inherit the permissions they are granted. When manufacturers lack visibility into sensitive engineering files and production data, they also lose visibility into what AI tools and non-human identities can reach.
AI Is Expanding the Manufacturing Attack Surface
Manufacturing has used automation for decades. AI extends it further: predictive maintenance platforms analyze equipment telemetry continuously, copilots help engineers write code and review designs, and supply chain tools exchange data directly with suppliers and logistics providers.
Each of those capabilities introduce identities that need authenticated access to enterprise resources. The Netwrix report found that 58 percent of organizations say AI has increased the number of identities with access to enterprise data, 72 percent report that identity-related exposure risk has increased, and 41 percent already run agentic AI in production with access to enterprise information. At the same time, 76 percent of organizations do not fully govern or monitor non-human identities.
In manufacturing, those identities extend well beyond corporate IT. Machine identities on MES platforms, programmable logic controllers, industrial IoT devices, and robotics platforms interact with enterprise data, and they were never designed for modern identity governance programs. Rotating one of them can require a maintenance window, which is part of why they persist unchanged.
Manufacturing's Identity Governance Is Falling Behind
Only 19 percent of manufacturing respondents say they are fully confident they can detect risky or toxic combinations of access, compared to 25 percent across all industries. Just 24 percent have implemented just-in-time privileged access, compared to 30 percent overall.
Third-party access is handled even less formally, with 54 percent of manufacturers managing privileged vendor access through manual, case-by-case processes, compared to 49 percent across all industries.
That matters in environments where equipment vendors, systems integrators, and maintenance contractors need privileged access to production systems. Manual processes increase the likelihood that privileged access remains active longer than necessary or escapes regular review.
Legacy Identity Infrastructure Still Creates Risk
Manufacturing identity infrastructure has grown over decades, through acquisitions, facility expansions, and the gradual convergence of IT and operational technology. Hybrid Active Directory deployments in those environments support both legacy production systems and modern cloud services.
Across all industries, only 26 percent of organizations are fully confident their Active Directory environment is free of misconfigurations that could enable privilege escalation. Among manufacturers, the figure is 20 percent.
AI systems authenticate through the same certificates, service accounts, and directory services. Weaknesses that once affected only employee accounts can now extend to AI agents and automated workflows operating continuously across production environments.
Governance Has Not Kept Pace with Adoption
The obstacle in front of manufacturers is not AI adoption itself but governance. Only 18 percent of manufacturing organizations report fully governing and continuously monitoring non-human identities, compared to 24 percent across all industries. A dedicated AI governance function or leader exists at 13 percent.
Visibility into AI itself remains limited. Only 13 percent of manufacturers report complete visibility into what sensitive data is being used by AI tools, models, or copilots, compared to 21 percent across all industries. Fully governing employee use of unsanctioned or shadow AI tools is reported by 15 percent, versus 24 percent overall.
Across all industries, budget constraints remain the primary obstacle preventing organizations from improving identity and data security. For manufacturing organizations, the largest barrier is skills and staffing shortages.
Securing a modern manufacturing environment calls for expertise in identity governance, operational technology, cloud platforms, and non-human identities. That combination is what makes the role hard to fill.
Where to Start
For years, manufacturers have focused cybersecurity investments on protecting production systems from disruption. That objective remains essential, but AI is changing where risk originates.
Full discovery across a plant estate built over decades is a multi-year program. The first pass should cover the systems whose loss halts production or hands over the design: process recipes, CAD and PLM repositories, controls code, and quality records. Map which identities can reach each one before deciding what any AI tool is allowed to read.
From there, a few steps close the specific gaps above:
- Register every AI deployment the way a new employee is registered, with a named owner, a defined scope of readable data, prohibited actions, and an access expiry date.
- Discover service accounts, tokens, and certificates across both IT and OT, and attest to them on a fixed calendar instead of at project close. Anything without a current owner is a retirement candidate.
- Replace standing vendor privilege with time-boxed, approved sessions. Set zero standing privilege on production systems as the target, and route credential rotations that need a maintenance window through the change process that schedules mechanical work.
- Run an Active Directory attack path assessment before connecting more agents to the directory. Every AI workload authenticating through AD extends whatever weakness is already there.
- Give AI-related identity risk one named owner.
As AI becomes embedded across modern manufacturing, identity security moves beyond the data center and onto the factory floor.
Darryl Baker is a Senior Staff Security Researcher at Netwrix, where he focuses on identity security and emerging attack techniques.
