How to implement AI in OT: a practical guide for industrial plants (2026)

Updated: 24 October 2026 · SIM Solutions

Short answer

Pick two or three tasks where operators, technicians or engineers lose time every day. Give the AI read-only access to the data those tasks need, such as alarm history, trends, manuals and PLC or SCADA projects. Prove it on one line with success measures agreed in advance, keep a person in control of every change, and only then roll out.

What "AI in OT" means in practice

Most plants already use some form of AI, such as predictive models on one machine. What changed is language models: assistants that read alarms, trends, manuals and control software, answer questions in plain language, and draft documents and code. In OT they are useful in two ways:

Assistants that read and explain. They find the first cause of a stop, explain an alarm, search manuals and history, and draft handovers and work orders. Low risk and quick to prove.
Agents that act. They write PLC code, build SCADA screens or change configurations. Powerful, but only on development or test systems, with an engineer approving each change.

Vendors are moving fast too. In June 2026, for example, Siemens announced an engineering agent for TIA Portal that writes PLC code and builds HMI screens. For most plants the question is no longer whether AI can do this, but how to bring it in safely across all the systems they already run.

Step 1: pick use cases that pay off

Good first use cases share four things: they happen often, they cost time or production, the data already exists, and a wrong answer cannot hurt anyone because a person still decides.

Use case	Data it needs	Risk
First cause of a line stop	Alarm and event history, interlock logic	Low: the operator decides
Alarm explanations and procedures	Alarm texts, operating procedures	Low
Troubleshooting support	Trends, work order history, manuals	Low: the technician decides
Shift handover and reports	Production log, alarms, setpoint changes	Low
PLC and SCADA documentation	PLC and SCADA project files	Low: read-only
Legacy SCADA or PLC migration	Old project files, tag lists, standards	Medium: done on a development system, then tested
Code for new logic	Company standard, block library	Medium: engineer review and simulation tests

Step 2: get data access right

The assistant is only as good as the data it can see, and in OT that access must not create a new risk.

Read from copies or read-only interfaces: the historian, the alarm database, exported project files, CMMS exports.
Keep control networks segmented. Connect through the DMZ or a read-only gateway, never with a direct path from the internet to a PLC.
Give the assistant its own account with the fewest rights it needs, and log every request.
Never let a language model write to PLCs or SCADA in production. Changes go through engineering, testing and approval.

Step 3: choose where the models run

Three options are common, and many plants mix them:

Enterprise AI services from Anthropic or OpenAI. Their business terms state that your data is not used for training by default.
The same models in EU cloud regions, for example through Amazon Bedrock or Google Cloud, when data residency matters.
Local models on site for data that may not leave the plant, usually with less capability.

Step 4: pilot on one line

Agree on success measures before the start, for example how long it takes to find the first cause of a stop, how long a handover takes, or how many hours a migration needs. Then let operators and engineers use the assistant in their real work for a few weeks. Their feedback decides what goes to the next line.

Step 5: governance and security

In December 2025, nine cyber security agencies, including the US CISA, Germany's BSI, the Dutch NCSC and the UK NCSC, published joint guidance on integrating AI into OT. It sets out four principles:

Understand AI. Know its specific risks in OT, train your people, and use a secure development lifecycle.
Consider AI use in the OT domain. Check the business case, manage OT data security risks, and account for vendors and integration.
Establish governance and assurance. Bring AI under your existing security frameworks, test and evaluate models continuously, and track regulation.
Embed safety and security. Keep human oversight, keep systems transparent, and include AI in incident response.

The EU AI Act: what applies when

The AI Omnibus entered into force on 27 July 2026 and moved the high-risk deadlines: 2 December 2027 for the stand-alone high-risk systems in Annex III, and 2 August 2028 for AI built into products under EU product law. Most plant assistants that explain or draft are not high-risk, but AI that acts as a safety component can be. Check each use case early; for a legal opinion, ask a lawyer.

Common mistakes

Starting with a general chatbot that cannot see plant data, so the answers stay generic.
Giving write access to control systems too early.
Building it without operators and technicians, who then do not use it.
No owner after the pilot, so it never reaches the second line.

How we help

We are automation engineers who build AI. We assess use cases, connect AI safely to your PLC, SCADA, historian and maintenance data, run the pilot and roll it out, in IT and OT. See AI implementation and the demos.

Questions we get

How soon do we see results?

A read-only assistant on one line can be tested quickly, because it uses data that already exists. Engineering use cases such as migrations take longer, because every result is tested.

Do we need a data platform or Unified Namespace first?

No. A good first use case reads the data that already exists. A Unified Namespace or data platform helps when you scale to many lines and sites.

Will AI replace operators or engineers?

No. It removes searching and writing work and gives faster answers. People keep the decisions, especially in OT.

Which systems can it connect to?

Most SCADA, historian and maintenance systems, through exports, databases or APIs. We have our own connectors for Siemens TIA Portal, WinCC V7 and Ignition.

How to implement AI in OT: a practical guide for industrial plants