REF / 90 terms

Glossary: AI governance, OT security and industrial data

Plain definitions of the terms you meet in IEC 62443, NIST, the EU AI Act, NIS2 and industrial data projects, written for a manufacturing context.

A

AI governance committeeAI governance#
A small cross-functional group that approves AI use cases, sets policy and owns AI risk decisions. In a mid-market manufacturer it typically includes operations, IT, OT or controls engineering, quality, EHS, legal or compliance and HR, with an executive sponsor. It meets on a fixed cadence, keeps the use case register current and escalates high-risk decisions to leadership. Read more about AI governance committee
AI impact assessmentAI governance#
A structured review of how an AI system could affect people, safety, product quality and the business before it is deployed or materially changed. It records intended use, foreseeable misuse, affected groups, data sources, failure modes and the controls in place. ISO/IEC 42001 expects organizations to assess AI system impacts; the EU AI Act separately requires a fundamental rights impact assessment from certain deployers of high-risk systems. Read more about AI impact assessment
AI literacyRegulation#
The skills, knowledge and understanding people need to use AI systems sensibly and to recognize their risks. Article 4 of the EU AI Act, as amended by the Digital Omnibus on AI in July 2026, requires providers and deployers to take measures to support the AI literacy of staff who operate or use AI systems on their behalf. For a plant, that means role-based training for operators, engineers and managers who rely on AI outputs. Read more about AI literacy
AI management systemAI governance#
The set of policies, roles, processes and records an organization uses to govern the development, provision or use of AI in a controlled, repeatable way. ISO/IEC 42001 specifies requirements for an AI management system (AIMS) using the same management system structure as ISO 9001 and ISO/IEC 27001, so manufacturers can integrate it with existing quality and security systems. Read more about AI management system
AI RMFAI governance#
The NIST Artificial Intelligence Risk Management Framework (AI RMF 1.0), published in January 2023. It is voluntary and organized around four functions: Govern, Map, Measure and Manage. NIST has also published a Generative AI Profile (NIST AI 600-1). Manufacturers use it as a practical vocabulary for identifying, assessing and treating AI risks, especially in the US where no single AI law applies. Read more about AI RMF
AI use case registerAI governance#
A living inventory of every AI system or AI-enabled tool the organization uses, builds or is piloting, with owner, purpose, data used, affected processes, vendor and risk tier. It is the starting point for AI governance: you cannot assess or control AI you do not know about. In a plant it should include AI embedded in vendor equipment and SaaS tools, not just in-house projects. Read more about AI use case register
AI-BOMAI governance#
An AI bill of materials: a structured record of what an AI system is built from, such as the model and its version, training and fine-tuning datasets, libraries, pretrained components and external services. It extends the SBOM idea to AI. For a manufacturer buying a vision or predictive maintenance system, asking for an AI-BOM shows where models and data come from and what must be re-checked when the vendor updates them. Read more about AI-BOM
Anomaly detectionAI governance#
An analytics or machine learning technique that flags data points or patterns that differ from normal behavior. In manufacturing it is used on sensor and process data to spot early equipment faults, process drift or quality issues, and on OT network traffic to spot unusual connections. Its outputs are alerts that need a human to interpret them, and false positive rates matter for alarm fatigue. Read more about Anomaly detection
Asset ownerOT security#
In IEC 62443, the organization accountable for an industrial automation and control system and responsible for operating it securely. For a manufacturer this is usually the plant or company itself, even when an OEM built the machine or an integrator maintains it. The asset owner sets target security levels, defines zones and conduits and decides which risks are accepted. Read more about Asset owner

B

BrownfieldOT security#
An existing plant, line or control system that was installed years ago and must keep running while it is improved. Brownfield environments often include unsupported operating systems, PLCs that cannot be patched quickly, undocumented network connections and OEM-owned equipment. Security and AI projects in brownfield sites need compensating controls, careful maintenance windows and management of change rather than rip-and-replace. Read more about Brownfield

C

CE markingRegulation#
The mark a manufacturer affixes to a product to declare that it conforms to all applicable EU harmonisation legislation, such as the Machinery Regulation, Low Voltage Directive or Cyber Resilience Act. It requires a conformity assessment, technical documentation and an EU declaration of conformity. AI components and cybersecurity features can now affect whether machinery and connected products can carry the CE mark. Read more about CE marking
CMMSData#
Computerized maintenance management system: software that manages work orders, preventive maintenance schedules, spare parts and asset maintenance history. In a predictive maintenance project the CMMS is both a data source, because failure and repair history labels the sensor data, and a destination, because model alerts should become work orders. Poor failure coding in the CMMS is a common reason models underperform. Read more about CMMS
Computer vision QCAI governance#
Automated quality inspection that uses cameras and image analysis, often machine learning, to detect defects, check assembly or measure parts on the line. It can inspect every part at line speed but depends on lighting, labeled defect images and stable product variants. Where results drive product release or safety-relevant decisions, it needs validation, human oversight and change control like any other quality instrument. Read more about Computer vision QC
Condition monitoringData#
Continuous or periodic measurement of equipment health indicators such as vibration, temperature, current draw, oil quality or acoustic signals. It shows the current state of an asset and triggers action when values cross thresholds. Condition monitoring is the data foundation for predictive maintenance, which adds models to estimate how condition will change and when failure is likely. Read more about Condition monitoring
ConduitOT security#
In IEC 62443, a logical grouping of communication channels that connect two or more zones and share the same security requirements. Examples include the path between the control zone and the site DMZ, or a vendor remote access path. Conduits are where you apply controls such as firewalls, protocol filtering, data diodes and authentication, so every allowed data flow should be documented as part of a conduit. Read more about Conduit
Conformity assessmentRegulation#
The process that demonstrates a product or system meets the requirements of EU legislation before it is placed on the market. Depending on the law and product category it may be internal control by the manufacturer or involve a notified body. Under the EU AI Act, high-risk AI systems need a conformity assessment, and since the 2026 Digital Omnibus, AI requirements for safety components of machinery are to be assessed through the Machinery Regulation's own conformity procedure. Read more about Conformity assessment
ContextualizationData#
Adding the meaning that raw OT data lacks: which asset, line, product, batch, units of measure, operating state and shift a value belongs to. A historian tag called FT_301.PV is useless to a data scientist without context. Contextualization is usually done with an asset model based on ISA-95, a unified namespace or an industrial data platform, and it is often most of the effort in an AI data project. Read more about Contextualization
CRARegulation#
The EU Cyber Resilience Act, Regulation (EU) 2024/2847. It sets cybersecurity requirements for products with digital elements placed on the EU market, including secure design, vulnerability handling, security updates, an SBOM and reporting of actively exploited vulnerabilities and severe incidents. Obligations phase in over several years. Manufacturers that sell connected machines, controllers or software are directly affected as product manufacturers. Read more about CRA
CVEOT security#
Common Vulnerabilities and Exposures: a public catalog that gives each disclosed software or firmware vulnerability a unique identifier, such as CVE-2024-12345. The program is operated by MITRE with US government sponsorship. OT teams match CVEs against their asset inventory and vendor advisories to see which PLCs, HMIs and network devices are affected, then prioritize using exploitation data and the device's exposure. Read more about CVE

D

Data diodeOT security#
A unidirectional gateway that physically allows data to flow in only one direction, typically from the OT network out to IT or the cloud. Because no return path exists, it blocks inbound attacks through that connection. Data diodes are used to replicate historian data or send telemetry to analytics platforms from high-consequence zones, but they complicate protocols that need acknowledgments and remote support. Read more about Data diode
Data driftAI governance#
A change in the statistical properties of the input data an AI model receives compared with the data it was trained on. In a plant it can come from a new raw material supplier, a recalibrated sensor, a new product variant, seasonal ambient conditions or a control logic change. Data drift is a leading indicator that model performance may degrade, so inputs should be monitored after go-live. Read more about Data drift
Data poisoningAI governance#
An attack or failure in which training or reference data is manipulated or corrupted so that a model learns the wrong behavior. In manufacturing it could be mislabeled defect images, tampered sensor history or altered maintenance records. Controls include restricting who can change training data, versioning datasets, checking data provenance and validating models against a trusted holdout set before release. Read more about Data poisoning
Data qualityData#
How fit data is for its intended use, judged on dimensions such as completeness, accuracy, timeliness, consistency and validity. Industrial data has its own problems: flatlined sensors, compression that hides spikes, unsynchronized timestamps, manual entries and tags renamed after a controls change. Measuring data quality before an AI project avoids months of modeling on data that cannot support the use case. Read more about Data quality
DCSOT security#
Distributed control system: an integrated control system, common in continuous and batch process industries such as chemicals, food and pharmaceuticals, in which controllers distributed around the plant are managed from a central engineering and operator environment. A DCS usually comes from a single vendor with its own network, historian and HMI, which shapes what security tools and AI integrations are supported. Read more about DCS
Defense in depthOT security#
A security strategy that layers multiple independent controls so that no single failure exposes the process. In OT it combines physical security, network segmentation into zones, conduit controls, hardened endpoints, account management, monitoring and tested recovery. It is a core principle of IEC 62443 and NIST SP 800-82 and matters because many plant devices cannot be patched or run endpoint agents. Read more about Defense in depth
DeployerRegulation#
Under the EU AI Act, a natural or legal person that uses an AI system under its authority, except for purely personal non-professional activity. A manufacturer that buys and runs an AI vision system, scheduling tool or chatbot is a deployer. Deployers of high-risk systems have obligations such as following instructions for use, assigning human oversight, monitoring operation and keeping logs. Read more about Deployer
Digital twinData#
A digital representation of a physical asset, process or system that is kept synchronized with its real-world counterpart through data, and used to monitor, simulate or optimize it. In manufacturing it ranges from a 3D model linked to live machine data to a process simulation used for virtual commissioning. The term is used loosely, so ask what data, fidelity and update frequency a vendor actually means. Read more about Digital twin

E

Edge computingData#
Running data processing and AI inference on hardware located close to the machines, such as an industrial PC on the line or a gateway in the plant, rather than in a remote cloud. It reduces latency and bandwidth, keeps operating if the WAN fails and can keep sensitive process data on site. Edge devices still need patching, hardening and placement in the right OT zone. Read more about Edge computing
Engineering workstationOT security#
A computer used by controls engineers to program PLCs, configure HMIs and drives and download logic to devices. It holds project files and the software that can change the physical process, which makes it one of the most valuable targets in an OT network. It should sit in a protected zone, be tightly access controlled and not be used for email or web browsing. Read more about Engineering workstation
ERPData#
Enterprise resource planning: the business system that manages orders, inventory, purchasing, finance and production planning, sitting at Level 4 of the Purdue model and ISA-95. ERP sends production orders down to MES or the plant and receives confirmations back. AI use cases such as demand forecasting and scheduling often combine ERP data with shop-floor data, which creates new IT/OT data flows to govern. Read more about ERP
Essential entityRegulation#
A category of organization under NIS2 subject to the stricter supervisory regime, including proactive supervision and higher maximum fines. Essential entities are generally large organizations in the sectors of high criticality listed in Annex I, plus certain entities designated by member states. Most manufacturers fall under Annex II as important entities, but some, such as those in energy or water, can be essential. Read more about Essential entity
EU AI ActRegulation#
Regulation (EU) 2024/1689, the EU's horizontal law on artificial intelligence. It takes a risk-based approach: certain practices are prohibited, high-risk AI systems face requirements on risk management, data, documentation, human oversight and accuracy, some systems carry transparency duties and general-purpose AI models have their own obligations. Its obligations apply in phases, and the timeline for high-risk rules has been subject to amendment proposals. Read more about EU AI Act

F

FAT/SATOT security#
Factory acceptance test and site acceptance test. The FAT verifies a machine or control system at the supplier's facility before shipment; the SAT verifies it after installation at the plant, connected to real utilities and networks. Both are the right moments to check security requirements, such as default passwords removed, unused services disabled, remote access as agreed and documentation of network connections. Read more about FAT/SAT
Foundational requirementsOT security#
The seven groups of security requirements in IEC 62443: FR 1 identification and authentication control, FR 2 use control, FR 3 system integrity, FR 4 data confidentiality, FR 5 restricted data flow, FR 6 timely response to events and FR 7 resource availability. System and component requirements in IEC 62443-3-3 and 4-2 are organized under these FRs, and security levels are expressed per FR. Read more about Foundational requirements

G

Generative AIAI governance#
AI that produces new content such as text, code, images or summaries in response to prompts, usually built on large language models or similar foundation models. In manufacturing it is used for drafting work instructions, summarizing maintenance logs, troubleshooting help and code assistance. Its main risks are confidential data leaving the company, plausible but wrong answers and unreviewed outputs reaching the shop floor. Read more about Generative AI
GPAI modelRegulation#
General-purpose AI model: under the EU AI Act, an AI model trained on large amounts of data that displays significant generality and can competently perform a wide range of distinct tasks, and can be integrated into many downstream systems. Large language models are the typical example. GPAI obligations fall on the model provider, not on a manufacturer that simply uses a product built on one. Read more about GPAI model

H

HallucinationAI governance#
Output from a generative AI model that is fluent and confident but factually wrong or unsupported, such as an invented torque value, a non-existent standard clause or PLC code that calls a function the platform does not have. Hallucinations are a property of how these models work, not a rare bug. Any generative AI output that affects equipment, safety or quality needs human review against a trusted source. Read more about Hallucination
High-risk AI systemRegulation#
Under the EU AI Act, an AI system that is either a safety component of, or itself, a product covered by EU harmonisation legislation listed in Annex I that requires third-party conformity assessment, such as certain lifts, pressure equipment or medical devices (machinery follows a separate route since the 2026 Digital Omnibus), or that falls into a use case listed in Annex III, such as some employment or critical infrastructure uses. High-risk systems carry the Act's heaviest requirements for providers and deployers. Read more about High-risk AI system
HistorianData#
A database optimized for storing and retrieving large volumes of time-stamped process data from PLCs, DCS and SCADA, such as temperatures, pressures, speeds and states. It usually sits at Level 3 of the Purdue model, with a replica often placed in the DMZ for business users. The historian is typically the first data source for analytics and AI, so how it connects to IT or the cloud is a key security decision. Read more about Historian
HMIOT security#
Human-machine interface: the screen or panel operators use to monitor and control a machine or process, showing values, alarms and trends and providing buttons and setpoints. HMIs range from small touch panels on a single machine to SCADA workstations covering a whole plant. Many run Windows-based software, so they need the same attention to accounts, patching and remote access as other OT computers. Read more about HMI
Human oversightAI governance#
The measures that let competent people understand, monitor and, where needed, override or stop an AI system. The EU AI Act requires high-risk systems to be designed for effective human oversight and deployers to assign it to trained staff. In a plant this means clear authority to reject an AI recommendation, visible confidence or reason information, and procedures that avoid automation bias. Read more about Human oversight

I

ICSOT security#
Industrial control system: a general term for the systems that monitor and control physical industrial processes, including PLCs, DCS, SCADA, HMIs, safety systems and the networks that connect them. ICS is often used interchangeably with OT, although OT also covers building systems and other equipment that interacts with the physical world. IEC 62443 uses the term industrial automation and control system (IACS). Read more about ICS
IDMZOT security#
Industrial demilitarized zone: a buffer network between the enterprise IT network and the plant OT network, often called Level 3.5 in Purdue model diagrams. No traffic should pass directly between IT and OT; instead, services such as historian replicas, patch servers, file transfer and remote access gateways sit in the IDMZ and broker the exchange. It is the natural place to control new AI and cloud data flows. Read more about IDMZ
IEC 62443OT security#
A series of international standards for the security of industrial automation and control systems, developed by the ISA99 committee and the IEC. It covers asset owner security programs, risk assessment with zones and conduits, system security requirements, secure product development and component requirements, and security levels. It assigns responsibilities to asset owners, integrators and product suppliers, which makes it useful for vendor contracts. Read more about IEC 62443
IIoTData#
Industrial Internet of Things: networked sensors, instruments, gateways and machines that collect and share industrial data, often for monitoring, analytics and AI. IIoT devices are frequently added to brownfield lines to capture data that existing PLCs do not expose, such as vibration or energy use. Each device is also a new networked asset that must be inventoried, secured and placed in the right zone. Read more about IIoT
Important entityRegulation#
A category of organization under NIS2 that must meet the same cybersecurity risk management and incident reporting obligations as essential entities, but is supervised reactively, after evidence of non-compliance, with lower maximum fines. Medium and large manufacturers of products such as machinery, electrical equipment, electronics, vehicles and medical devices are typically important entities under Annex II, subject to national transposition. Read more about Important entity
ISA-95Data#
An international standard, also published as IEC 62264, for integrating enterprise and control systems. It defines a functional hierarchy of levels, from business planning down to control, and standard models for information exchanged between ERP and manufacturing operations, such as production schedules, performance, materials and equipment. ISA-95 hierarchies are widely used to structure a unified namespace and AI data models. Read more about ISA-95
ISO/IEC 27001Regulation#
The international standard for information security management systems (ISMS), covering how an organization establishes, runs and improves security controls through risk assessment and an Annex A control set. Many manufacturers hold or are asked for ISO/IEC 27001 certification for IT. It does not address OT specifics such as safety and availability in depth, which is why it is commonly paired with IEC 62443. Read more about ISO/IEC 27001
ISO/IEC 42001AI governance#
ISO/IEC 42001:2023, the first international management system standard for artificial intelligence. It specifies requirements for establishing, implementing, maintaining and improving an AI management system, including AI policy, roles, risk and impact assessment, and controls in Annex A. It is certifiable and is designed to integrate with ISO 9001 and ISO/IEC 27001 management systems many manufacturers already run. Read more about ISO/IEC 42001

J

Jump hostOT security#
A hardened server, usually in the IDMZ, that remote users must connect to first before reaching OT systems. It concentrates authentication, multi-factor login, session logging and access approval in one controlled point instead of allowing direct connections into the plant. A jump host is a common building block of secure remote access for OEMs and integrators, ideally with time-limited, approved sessions. Read more about Jump host

K

KEVOT security#
The Known Exploited Vulnerabilities catalog maintained by the US Cybersecurity and Infrastructure Security Agency (CISA). It lists CVEs with reliable evidence of active exploitation. Because most plants cannot patch everything, KEV is a practical filter: a vulnerability on an exposed OT or IDMZ asset that appears in KEV should move to the top of the remediation or mitigation list. Read more about KEV

L

Large language modelAI governance#
A machine learning model trained on very large amounts of text to predict and generate language, used in chat assistants, copilots and document tools. LLMs can summarize, translate, draft and write code, but they do not know your plant unless given your data and they can produce confident errors. Under the EU AI Act most large language models are general-purpose AI models. Read more about Large language model
Least privilegeOT security#
The principle that every user, service account, device and application gets only the access it needs to do its job, for only as long as it needs it. In OT that means separate operator, engineer and administrator roles, no shared admin accounts on HMIs, vendor access limited to specific machines and AI services reading data without write access to controllers. Read more about Least privilege

M

Machinery RegulationRegulation#
Regulation (EU) 2023/1230, which replaces the Machinery Directive 2006/42/EC. It updates essential health and safety requirements, including protection against corruption, cybersecurity-related safety risks and machinery with self-evolving behavior, and requires third-party assessment for some categories. It matters for manufacturers that build machines and for those adding AI safety functions to equipment. Read more about Machinery Regulation
MESData#
Manufacturing execution system: software at Level 3 of the Purdue model and ISA-95 that manages and records production on the shop floor, including work order execution, genealogy and traceability, quality data, labor and performance tracking. MES connects ERP plans with real machine activity. It is a rich source of contextual data for AI, linking process data to products, batches and orders. Read more about MES
MFAOT security#
Multi-factor authentication: requiring two or more independent proofs of identity, such as a password plus an authenticator app or hardware token, before granting access. In OT it is most important on remote access paths, jump hosts, the IDMZ and privileged accounts. Many controllers and HMIs cannot support it directly, so it is usually enforced at the conduit or gateway in front of them. Read more about MFA
MOCOT security#
Management of change: the formal process used in plants to review, approve, document and communicate changes to equipment, processes, procedures and control systems before they are made. It originated in process safety. Adding AI to a process, connecting a new data flow or changing model logic that influences operations should go through MOC so safety, quality and security impacts are checked. Read more about MOC
Model driftAI governance#
The decline in an AI model's performance over time as the real world it operates in changes, also called model decay. It can be caused by data drift or by a change in the relationship between inputs and outcomes, such as a new failure mode after a component change. Detecting it requires tracking model accuracy against actual outcomes and defining when to retrain or roll back. Read more about Model drift
MQTTData#
A lightweight publish-subscribe messaging protocol in which devices publish messages to topics on a broker and other applications subscribe to them. It is an OASIS standard and is widely used in IIoT and unified namespace architectures because it is efficient and decouples data producers from consumers. Brokers need authentication, TLS and topic-level access control, since by default MQTT trusts whoever connects. Read more about MQTT

N

Network segmentationOT security#
Dividing a network into separate sections with controlled communication between them, using firewalls, VLANs with access controls or physical separation. In OT it limits how far ransomware or an intruder can spread and keeps business traffic away from control traffic. IEC 62443 formalizes segmentation as zones and conduits, and it is usually the highest-value security investment for a mid-market plant. Read more about Network segmentation
NIS2Regulation#
Directive (EU) 2022/2555 on measures for a high common level of cybersecurity across the Union. It expands the sectors and entities covered compared with the original NIS Directive, including many manufacturing subsectors, and requires cybersecurity risk management measures, management body accountability, supply chain security and staged incident reporting. As a directive, it applies through each member state's national law. Read more about NIS2
NIST CSFRegulation#
The NIST Cybersecurity Framework, a voluntary framework for managing cybersecurity risk. Version 2.0, released in February 2024, organizes outcomes into six functions: Govern, Identify, Protect, Detect, Respond and Recover. It is widely used by US manufacturers and is a practical way to present one cybersecurity program that spans IT and OT to executives and boards. Read more about NIST CSF
NIST SP 800-82OT security#
NIST Special Publication 800-82, the Guide to Operational Technology Security. Revision 3, published in September 2023, broadened the scope from industrial control systems to OT generally and provides an OT overlay of NIST SP 800-53 controls. It is a free, detailed reference for OT architecture, threats and countermeasures that pairs well with IEC 62443. Read more about NIST SP 800-82
Notified bodyRegulation#
An organization designated by an EU member state to carry out third-party conformity assessment of products under specific EU legislation before they can be placed on the market. Notified bodies are listed in the Commission's NANDO database. Whether you need one depends on the law and product category, for example certain machinery, high-risk AI systems or critical products with digital elements. Read more about Notified body

O

OEEData#
Overall equipment effectiveness: a manufacturing metric equal to availability multiplied by performance multiplied by quality. It shows what share of planned production time is truly productive. OEE is a common target for AI projects, from predictive maintenance to reduce downtime to vision inspection to reduce scrap, and it gives a business baseline to measure whether an AI pilot paid off. Read more about OEE
OPC UAData#
OPC Unified Architecture: a platform-independent industrial communication standard, published as IEC 62541, for exchanging data and information models between devices and applications. It supports rich semantic models and built-in security features including authentication, signing and encryption, which must be configured rather than left in a no-security mode. It is a common way to get structured data out of PLCs and machines. Read more about OPC UA
OTOT security#
Operational technology: hardware and software that monitors or controls physical equipment, processes and events, such as PLCs, DCS, SCADA, HMIs, robots, drives and safety systems. OT priorities differ from IT: availability and safety usually come first, systems run for decades and changes need maintenance windows. Most AI in manufacturing depends on OT data, which is why OT security and AI governance meet. Read more about OT
OT asset inventoryOT security#
A maintained list of OT devices and software, with make, model, firmware, location, zone, owner, network connections and criticality. It is the foundation for vulnerability management, segmentation, incident response and NIS2 or IEC 62443 programs. It can be built from passive network monitoring, vendor documentation and walkdowns, and it should include AI edge devices and data gateways. Read more about OT asset inventory

P

PLCOT security#
Programmable logic controller: a ruggedized industrial computer that reads inputs from sensors, executes control logic and drives outputs such as motors, valves and actuators in real time. PLCs run machines and lines for decades, often cannot be patched without downtime and frequently lack strong authentication. Any AI system or data platform that can write to a PLC needs careful review. Read more about PLC
Predictive maintenanceData#
Maintenance planned from data and models that estimate when equipment is likely to fail, rather than on a fixed schedule or after breakdown. It uses condition monitoring data, process data and maintenance history to detect degradation early. It works best on critical assets with recurring, data-visible failure modes and needs a path from model alert to planned work order to deliver value. Read more about Predictive maintenance
Product supplierOT security#
In IEC 62443, the manufacturer of automation components or systems, such as PLCs, network devices, HMI software or complete control systems. IEC 62443-4-1 covers the supplier's secure development lifecycle and 4-2 the technical security capabilities of components. Asking product suppliers for 62443 certification or evidence helps an asset owner understand what security a product can actually provide. Read more about Product supplier
Product with digital elementsRegulation#
The CRA term for a software or hardware product and its remote data processing solutions, including software or hardware components placed on the market separately. It covers a wide range of connected products, from controllers and gateways to machine software. Products already covered by certain sector rules, such as medical devices or motor vehicles, are excluded, so check scope carefully. Read more about Product with digital elements
Prohibited AI practicesRegulation#
AI uses banned outright under Article 5 of the EU AI Act, such as manipulative techniques that cause significant harm, social scoring and, with narrow exceptions, emotion recognition in the workplace. The workplace emotion recognition ban is the one most relevant to manufacturers, for example in operator fatigue or engagement monitoring products that infer emotions. Check vendor features against Article 5 before deployment. Read more about Prohibited AI practices
Prompt injectionAI governance#
An attack on systems built with large language models in which crafted text, placed in a user prompt or hidden in a document, web page or email the model reads, causes the model to ignore its instructions or take unintended actions. It matters most when an AI assistant can call tools or access internal systems. Limit what AI agents can do and keep them away from control actions. Read more about Prompt injection
ProviderRegulation#
Under the EU AI Act, a person or organization that develops an AI system or general-purpose AI model, or has one developed, and places it on the market or puts it into service under its own name or trademark. Providers carry most obligations for high-risk systems. A manufacturer can become a provider by building AI into machines it sells or substantially modifying a high-risk system. Read more about Provider
Purdue modelOT security#
A reference model for industrial network architecture that arranges systems into levels: Level 0 physical process, Level 1 basic control such as PLCs, Level 2 supervisory control such as HMI and SCADA, Level 3 site operations such as MES and historian, and Levels 4 and 5 enterprise IT, with an industrial DMZ between 3 and 4. It remains a common way to discuss segmentation, even as cloud and IIoT blur the levels. Read more about Purdue model

R

RACIAI governance#
A responsibility matrix that shows who is Responsible, Accountable, Consulted and Informed for each activity. For AI governance in a plant it clarifies who approves a use case, who owns the model in production, who handles OT changes, who signs off quality impacts and who talks to vendors. One Accountable person per activity avoids decisions falling between IT, OT and operations. Read more about RACI
RansomwareOT security#
Malicious software that encrypts or steals data and demands payment to restore access or prevent publication. Manufacturers are frequent targets because downtime is expensive. Many plant shutdowns happen not because ransomware reached PLCs but because shared IT services, flat networks or a cautious disconnect stopped production. Segmentation, offline backups and practiced recovery are the main defenses. Read more about Ransomware
Retrieval-augmented generationAI governance#
A design, often called RAG, in which a generative AI system first searches a set of approved documents, such as manuals, SOPs and maintenance records, then answers using the retrieved text. It grounds answers in your own content and can cite sources, which reduces but does not eliminate hallucinations. Access controls on the document set matter, since the assistant can surface whatever it can read. Read more about Retrieval-augmented generation

S

SBOMOT security#
Software bill of materials: a machine-readable list of the software components, libraries and versions inside a product, commonly in SPDX or CycloneDX format. It lets an asset owner check quickly whether a newly disclosed vulnerability affects equipment on the floor. The Cyber Resilience Act requires manufacturers of products with digital elements to draw up an SBOM as part of their technical documentation. Read more about SBOM
SCADAOT security#
Supervisory control and data acquisition: systems that collect data from PLCs, RTUs and field devices and give operators supervisory monitoring and control over a process, often across a large plant or geographically spread assets. SCADA includes servers, HMI clients, communication links and alarm handling. It sits at Level 2 of the Purdue model and is a common source of data for historians and AI. Read more about SCADA
Secure remote accessOT security#
Controlled, auditable ways for employees, OEMs and integrators to connect to OT systems from outside the plant. Good practice includes a single brokered entry point such as a jump host in the IDMZ, individual accounts with MFA, access approved per session and time-limited, session recording and no always-on vendor VPNs or modems. It is one of the most common entry points for attackers. Read more about Secure remote access
Security levelOT security#
In IEC 62443, a measure of the security strength a zone, conduit or component needs or provides, on a scale from SL 0 (no special protection) to SL 4 (protection against intentional violation using sophisticated means with extended resources). Levels are stated per foundational requirement. Most mid-market plant zones target SL 1 or SL 2, with higher targets for critical or exposed zones. Read more about Security level
Shadow AIAI governance#
AI tools and features used inside the organization without approval, visibility or governance, such as engineers pasting PLC code or drawings into a public chatbot, quality staff using a free image tool or a vendor quietly adding AI to a product update. It is rarely malicious. The fix is a clear policy, an approved alternative and a simple way to register use cases, not just blocking. Read more about Shadow AI
SISOT security#
Safety instrumented system: an independent control system that takes a process to a safe state when dangerous conditions are detected, designed to functional safety standards such as IEC 61511 and IEC 61508. It is separate from the basic process control system. AI and analytics should never be allowed to write to or bypass an SIS, and SIS networks belong in their own highly protected zone. Read more about SIS
SL-T, SL-C and SL-AOT security#
The three types of security level in IEC 62443. SL-T, the target, is the level a zone or conduit needs based on risk assessment. SL-C, the capability, is the level a component or system can provide when properly configured. SL-A, achieved, is the level actually delivered in the installed system. Gaps between them show where compensating controls or upgrades are needed. Read more about SL-T, SL-C and SL-A
Sparkplug BData#
An open specification from the Eclipse Foundation, also published as ISO/IEC 20237:2023, that defines how industrial data is structured and transmitted over MQTT, including a topic namespace, payload encoding and device birth and death certificates that report state. It adds interoperability and state awareness that plain MQTT lacks, which is why it is often used in unified namespace and IIoT architectures. Read more about Sparkplug B

U

Unified namespaceData#
An architecture in which all systems publish their current data and events to a central, hierarchically organized structure, usually an MQTT broker using an ISA-95 style path such as site/area/line/cell. Consumers subscribe to what they need instead of building point-to-point integrations. It simplifies data access for AI but concentrates access, so the broker needs strong security and governance. Read more about Unified namespace

V

Vulnerability handlingRegulation#
A manufacturer's process for receiving, assessing, fixing and disclosing vulnerabilities in its products, including a contact point for reports, coordinated disclosure, security updates and advisories. The Cyber Resilience Act makes vulnerability handling a legal requirement for products with digital elements, and IEC 62443-4-1 describes it for automation suppliers. Asset owners should ask vendors how they do it. Read more about Vulnerability handling

Z

ZoneOT security#
In IEC 62443, a grouping of logical or physical assets that share common security requirements, such as one production line's controllers and HMIs, the safety system or the IDMZ. Each zone gets a target security level, and communication between zones flows only through defined conduits. Zoning is the main way to stop an incident in one area from spreading across the plant. Read more about Zone