For manufacturers, the AI Act does not sit in isolation. It is layered on top of longstanding product-safety regimes such as the Machinery Regulation, the Radio Equipment Directive, medical and automotive type-approval frameworks, and related harmonised standards. When AI performs safety functions or is integrated into regulated products, it triggers high-risk classification and associated obligations for risk management, documentation, conformity assessment, and post-market monitoring. At the same time, AI deployed on the factory floor for quality control, predictive maintenance, or worker-safety applications is drawn into a regulatory perimeter that manufacturers can no longer treat as purely internal or experimental.
This article examines how the AI Act will impact companies operating in the manufacturing industry. It focuses on the risk-based structure of the Act and its linkage to product-safety law, the industrial use cases most likely to be captured, the resulting obligations for manufacturers and their supply chains, and strategic responses at the level of governance, innovation and global market positioning.
1. The AI Act in the manufacturing context
1.1 Risk-based structure and product-safety linkages
The AI Act classifies AI systems into four categories: unacceptable-risk systems, which are prohibited; high-risk systems subject to stringent ex-ante controls; limited-risk systems with transparency obligations; and minimal-risk systems primarily governed by existing law and voluntary best practice. High-risk AI systems fall into two main groups: (a) systems that are safety components of products, or products themselves, covered by specified EU harmonisation legislation listed in Annex I; and (b) systems intended for high-risk use cases listed in Annex III, such as critical infrastructure management, employment and access to essential services.
For manufacturing, the Annex I route is central. The AI Act explicitly links to product-safety legislation governing machinery, low-voltage equipment, radio equipment, medical devices, in-vitro diagnostics, pressure equipment, and type-approval for vehicles and agricultural machinery. When an AI system performs a safety function within such a product and the product is subject to third-party conformity assessment, the AI system is classified as high-risk under Article 6(1). This design choice means that the regulatory treatment of AI in manufacturing is not divorced from the existing CE-marking landscape; it is integrated into it.
1.2 Definition and scope of safety components
The AI Act defines a “safety component” as a component of a product or of an AI system which fulfils a safety function, or whose failure or malfunctioning endangers the health and safety of persons or property. This definition is not identical to the concept of safety component in the Machinery Regulation, so manufacturers must assess safety functions under both instruments depending on whether they are analysing the machine as a whole or the AI system specifically.
In industrial practice, a wide range of AI systems may qualify as safety components: obstacle-detection and collision-avoidance algorithms for mobile robots; adaptive safety-rated speed and separation monitoring; AI-based emergency-stop logic; and systems that manage safe operating envelopes in collaborative robot applications. If such AI functionality can affect whether a machine stops in time, avoids human contact, or detects hazardous conditions, its failure can endanger workers or property and so falls squarely within the safety-component definition.
1.3 Timeline and transitional arrangements
The AI Act entered into force in 2024, with most obligations becoming applicable from 2 August 2026. However, for high-risk AI systems that are safety components of Annex I products, an extended transition period applies. The Commission’s guidance and sector-specific commentary indicate that full obligations for embedded high-risk AI in machinery and other Annex I products will phase in around 2 August 2027, with some proposals to extend certain deadlines to 2028 as part of an “AI Omnibus” simplification package.
Other elements come into force earlier. Prohibitions on unacceptable-risk practices apply from February 2025, and rules for general-purpose AI models and codes of practice start to apply in the period between 2025 and 2026. For manufacturers, whose product-development cycles often span several years, these timelines leave a relatively narrow window to redesign architectures, revise technical documentation and align quality-management systems so that AI components can pass through integrated conformity assessment alongside mechanical and electrical safety by the time grace periods expire.
2. Manufacturing use cases captured by the AI Act
2.1 AI as safety components of machinery and vehicles
AI systems that perform safety functions in machinery and vehicles are among the clearest candidates for high-risk classification. Article 6(1) provides that AI systems intended to be used as safety components of products covered by Annex I harmonisation legislation are high-risk if those products require third-party conformity assessment. This encompasses machinery under the Machinery Regulation, vehicles subject to type-approval, and agricultural and forestry machinery under Regulation (EU) 167/2013.
In the agricultural domain, guidance explains that AI systems performing safety functions on agricultural or forestry vehicles—such as lane-keeping, obstacle detection, roll-over prevention or automated braking—are high-risk because the vehicles themselves require third-party type-approval. AI therefore becomes part of the safety architecture that must be assessed by notified bodies. A similar logic applies to AI-based safety functions in construction machinery, lifting equipment and other machines subject to notified-body involvement under sectoral rules.
2.2 Industrial control, predictive maintenance and critical infrastructure
The AI Act also treats AI systems used in the management and operation of critical infrastructure as high-risk under Annex III. For manufacturing, this can include AI used in plant-wide control systems, process-control layers for chemical or energy-intensive operations, or supervisory systems that make or influence decisions about shutting down equipment, reconfiguring production lines, or responding to anomalies.
Predictive-maintenance applications illustrate the boundary issues. An AI system that predicts component failure and automatically adjusts operating parameters or triggers shutdowns in a way that materially affects safety may be treated as part of critical-infrastructure management and classified as high-risk. By contrast, a system that provides advisory predictions to human maintenance engineers, without direct control over equipment, may be treated as limited-risk but still subject to data-governance and documentation obligations. The distinction turns on the degree of autonomy, the safety significance of the decisions, and the integration with other safety functions.
2.3 Worker-safety, ergonomics and employment-related systems
Manufacturing plants increasingly use AI to protect workers: collision-avoidance systems in human-robot collaboration, real-time monitoring of unsafe postures or proximity to hazards, and dynamic safety zones that adjust based on worker location and behaviour. When these systems perform safety functions for machines or production lines, they will often be high-risk via the Annex I safety-component route.
Separately, the AI Act lists employment-related AI systems as high-risk when they are used for recruitment, selection, evaluation, or termination decisions. In manufacturing, this includes AI tools used to screen applicants for production roles, allocate shifts, predict absenteeism, or recommend disciplinary action based on production metrics or sensor-derived performance indicators. Even if such tools are deployed in HR rather than on the factory floor, they fall within the same regulatory framework and must meet high-risk requirements if they significantly influence employment decisions.
2.4 Quality control, inspection and process optimisation
AI-enabled quality control is a core industrial use case. Computer vision systems inspect surfaces, welds and assemblies; sensor-fusion models detect anomalies in process parameters; and multimodal systems assess product integrity on high-speed lines. Where these systems serve as safety components—because missed defects could endanger health, safety or the environment—they are likely to be high-risk.
Examples include AI systems that inspect welds in pressure vessels, detect flaws in structural components, or ensure that safety-critical features meet tolerances in braking systems or lifting equipment. If such inspection is integral to the product’s safety case and the machine requires third-party conformity assessment, the AI inspection module becomes part of the high-risk portfolio that must be documented and evaluated under the AI Act. Quality-control systems focused on cosmetic defects or non-safety-critical features may remain outside the high-risk category but can still be subject to limited-risk obligations related to transparency and documentation of AI use.
3. Obligations for manufacturing companies
3.1 Risk-management, data governance and documentation
Providers of high-risk AI systems must implement a formal risk-management system covering the entire lifecycle of the AI system, from design and development through testing, deployment and post-market monitoring. The process must identify risks associated with the intended use, estimate and evaluate their magnitude, and define control measures, with updates in response to new data and operational experience.
Data governance is a central pillar. The AI Act requires that training, validation and testing datasets be relevant, representative and free of significant errors, and that they be governed in a way that respects EU law, including non-discrimination and privacy. For industrial AI, this often means combining historical maintenance logs, sensor data from operational equipment, and simulation outputs, while documenting data sources, pre-processing, labelling procedures and quality-control steps. Poorly documented or biased datasets can undermine both safety and compliance.
Manufacturers must also produce detailed technical documentation and maintain records of the AI system and its performance. Documentation must describe the system’s design, algorithms, training process, risk controls, human-oversight mechanisms and expected performance characteristics in sufficient detail for authorities and notified bodies to assess compliance. High-risk AI systems must include automatic logging of key events, including inputs, outputs and relevant operating conditions, to support traceability and incident investigation.
3.2 Conformity assessment, CE marking and notified bodies
High-risk AI systems must undergo conformity assessment prior to being placed on the market or put into service. For AI embedded in machinery or other regulated products, the AI-specific assessment is intended to be integrated into the existing product-safety conformity procedures, rather than run as a separate process. This integration is vital in manufacturing, where products already go through CE-marking processes that address mechanical, electrical and electromagnetic safety.
Where sectoral legislation allows self-assessment, manufacturers can carry out the AI-related conformity assessment internally, provided they have appropriate quality-management systems in place. Where third-party assessment is required—such as for certain categories of machinery—AI-specific notified bodies will review the technical documentation, test results and risk-management processes related to the AI component. Certifications granted by notified bodies are time-limited, often expiring after five years, after which the system must be reassessed if it is still being marketed or has been significantly modified.
Once compliance is established, high-risk AI systems must bear the CE marking to signal conformity with the AI Act and, where applicable, with sectoral legislation. For machinery with embedded AI safety components, the machine’s CE mark must indicate compliance with both the Machinery framework and the AI Act. For purely digital AI systems, a digital CE mark may be used, with information provided in accompanying documentation or user interfaces.
3.3 Roles across the AI value chain
Manufacturing companies can be providers, deployers, importers and distributors under the AI Act. A machine builder that designs and integrates AI-based safety functions into its own machines is a provider of high-risk AI systems. The same company may also be a deployer when it uses AI for quality control, predictive maintenance or workforce management in its own plants.
Importers and distributors who place AI-enabled machinery on the EU market have their own duties: they must verify the presence of CE marking, ensure that declarations of conformity and instructions accompany the product, and refrain from placing systems on the market when they have reason to believe they are non-compliant. Non-EU manufacturers supplying the EU market will often appoint authorised representatives to handle documentation and regulatory interactions, mirroring arrangements under product-safety law.
3.4 Penalties, market surveillance and operational risk
The AI Act’s penalty regime is designed to be dissuasive. Serious infringements, including violations of prohibitions on unacceptable-risk practices or core obligations for high-risk AI systems, can attract fines of up to 35 million euros or 7 percent of global annual turnover, whichever is higher. Other major infringements can lead to fines up to 15 million euros or 3 percent of global turnover.
Beyond financial penalties, market-surveillance authorities can order withdrawal, recall or modification of non-compliant AI systems and can suspend or prohibit their use. For manufacturing companies, such measures can have immediate operational consequences: the suspension of a safety-critical AI system can halt production lines, delay shipments, and trigger contractual liabilities. This dynamic strengthens the case for integrating AI compliance into existing safety and quality-management systems, rather than treating it as an isolated legal requirement.
4. Operational impacts on factories and supply chains
4.1 Mapping AI systems and classifying risk
A foundational step for manufacturing companies is to compile a comprehensive inventory of AI systems used in their products and operations and to classify each system according to the AI Act’s risk categories. This mapping must cover embedded AI in machinery and components, standalone software controlling production processes, cloud-based AI services used by plants, and employment-related AI tools used by HR or operations departments.
Classification is not merely descriptive; it shapes governance obligations. Systems that qualify as high-risk through Annex I or Annex III require risk-management, documentation, conformity assessment and post-market monitoring. Limited-risk systems must meet transparency obligations and may still need documentation and data-governance controls. Minimal-risk systems can be governed primarily through internal policies, although they remain subject to general EU law. Conducting this mapping on a plant-by-plant and business-unit basis often reveals more AI use than management assumed, especially when third-party tools are embedded in vendor solutions.
4.2 Integrating AI into quality and safety management
The AI Act effectively brings AI into the same governance orbit as mechanical and electrical safety. For manufacturers, this means integrating AI-specific requirements into existing quality-management systems (QMS) and safety processes, rather than creating parallel structures. Design reviews should include AI risk assessments; factory acceptance tests must validate AI-based functions under realistic conditions; and change-control procedures must track modifications to models and data pipelines.
Training and competence are also affected. Safety engineers and operators need to understand how AI systems behave, under which conditions performance degrades, and how to interpret system outputs and logs. For example, operators must know when to override AI recommendations, and safety engineers must be able to interpret confusion matrices and performance metrics in the context of hazard analysis. This extends traditional safety engineering into probabilistic and data-driven territory.
4.3 Data infrastructure, logging and traceability
Industrial AI systems rely on operational data. The AI Act’s data-governance and logging requirements create an incentive for manufacturers to invest in data infrastructure that can support high-quality, auditable AI. This includes secure central data platforms, data-quality controls, metadata management and retention policies aligned with regulatory requirements and business needs.
Automatic logging of AI system behaviour is a key requirement for high-risk systems. Logs must capture relevant inputs, outputs, model versions, and contextual information about operating conditions, enabling post-incident analysis and continuous improvement. From an operational perspective, such logs can support root-cause analysis of failures and process deviations, but they also introduce cybersecurity obligations: logs themselves can be sensitive and must be protected against tampering, loss or unauthorised access.
4.4 Effects on suppliers and global supply chains
The AI Act’s obligations cascade through supply chains. OEMs that integrate AI-enabled components into machinery will expect their suppliers to provide technical documentation, risk assessments and declarations of conformity for AI subsystems, particularly when these subsystems perform safety functions. Suppliers that rely on opaque third-party AI modules without sufficient documentation may find themselves excluded from tenders or facing stringent contractual requirements.
Conversely, component manufacturers that invest early in AI compliance—standardising documentation, aligning with harmonised standards, and designing interfaces that support auditing—may gain a competitive advantage as OEMs seek to reduce friction in their own conformity-assessment processes. Non-EU suppliers that wish to maintain access to the EU market will need to align their practices with AI Act expectations, potentially using them as a baseline for global operations.
5. Strategic implications and industry responses
5.1 Balancing safety gains and compliance burdens
Industry associations and technical institutes acknowledge that AI can improve safety and productivity in manufacturing, but they also warn of overlapping and sometimes misaligned regulatory obligations. AI-based safety functions can reduce human error, detect hazardous conditions earlier, and support more responsive shutdowns, yet high-risk classification and duplicative conformity assessments under both product-safety law and the AI Act may slow deployment.
Some analyses argue that high-risk obligations should focus on applications where AI substantially alters the risk profile, rather than imposing full fundamental-rights assessments on every safety-related control. The development of harmonised standards will be critical in clarifying how risk management should be tailored to industrial use cases and in avoiding unnecessary duplication of tests already conducted under sectoral safety regimes.
5.2 Preparedness of non-EU manufacturers
Research on non-EU manufacturers’ preparedness suggests that companies in countries such as the United Kingdom and Brazil are unevenly aware of AI Act implications, even where they depend heavily on exports to the EU. Many firms treat AI functionality as a differentiating feature rather than as part of the regulated safety architecture, and they may lack internal processes for documenting AI systems at the level required by European regulators.
As the AI Act moves into the application phase, non-EU manufacturers will need to decide whether to treat EU-compliant AI design and documentation as a baseline for all markets or to maintain distinct product variants and processes for Europe. The experience with GDPR suggests that, over time, many globally active firms may converge on EU-aligned practices to avoid fragmentation and reduce regulatory uncertainty across jurisdictions.
5.3 Innovation, SMEs and industrial competitiveness
The AI Act’s impact on innovation and competitiveness in manufacturing is likely to vary by company size and sector. Large manufacturers often have established regulatory and quality-assurance capabilities and may integrate AI governance into existing structures with manageable incremental cost. Small and medium-sized enterprises (SMEs), particularly those that act as suppliers rather than OEMs, may find the fixed costs of risk assessments, documentation and notified-body interactions proportionally higher.
Policy initiatives, including regulatory sandboxes, guidance and funding instruments for AI innovation, are intended to alleviate some of these burdens for SMEs and to encourage the development of AI that is “compliance-by-design”. Nonetheless, there is a risk that complex compliance regimes could skew adoption towards larger players and slow diffusion of beneficial AI into the long tail of smaller manufacturers unless support mechanisms are effectively implemented.
5.4 Global regulatory spillovers and technical standard-setting
As a comprehensive and detailed AI regulation, the AI Act is likely to influence emerging frameworks in other jurisdictions and to shape technical standard-setting in industrial AI safety. Manufacturers that adapt early to AI Act requirements and engage in international standardisation bodies can help define practical norms for AI-based control systems, risk assessment and human-machine interaction.
This dynamic is visible in discussions around functional safety standards such as IEC 61508 and ISO 13849, which are being revisited to account for machine-learning components whose behaviour is not fully deterministic. The AI Act’s focus on risk management, data governance and human oversight provides a regulatory anchor for these technical developments and may accelerate convergence between legal requirements and engineering practice.
6. Governance recommendations for manufacturers
6.1 Build cross-functional AI governance
Manufacturers should establish cross-functional AI governance programs that bring together product-safety engineers, operational technology (OT) specialists, IT and cybersecurity teams, data scientists, legal and compliance functions, and business units. This structure should oversee AI system inventories, risk classification, conformity assessments and incident reporting, ensuring that AI-related risks are considered alongside mechanical, electrical and cyber risks.
Board-level oversight of AI is becoming a realistic expectation, especially in companies that operate critical infrastructure or supply safety-critical equipment. Integrating AI into enterprise risk-management frameworks can help decision-makers understand how AI affects safety, operational continuity, liability and reputation.
6.2 Align AI lifecycle with quality-management and product-development processes
Rather than building separate AI compliance tracks, manufacturers should align AI development and integration with existing quality-management and product-development processes. Design inputs should include AI-specific requirements; design reviews should cover AI model selection, training data, and risk controls; and validation testing should explicitly evaluate AI performance under representative and edge-case conditions.
Configuration and change management must extend to AI models. When models are retrained with new data or updated after deployment, manufacturers must assess whether these changes constitute “significant modification” requiring renewed conformity assessment or updated technical documentation. Clear procedures for model change control will be essential to avoid inadvertent non-compliance.
6.3 Invest in data, logging and traceability infrastructure
Complying with AI Act requirements on data governance and logging requires investments in data infrastructure. Manufacturers should develop secure, well-documented pipelines for collecting, storing and processing data used to train and operate AI systems, with controls to ensure data quality, provenance and access management.
Logging infrastructure should be designed to support both compliance and operational objectives. Comprehensive logs of AI system behaviour enable better diagnostics and continuous improvement, but they must be protected against tampering and unauthorised access, and retained for appropriate periods. Integrating logging with existing industrial control systems and cybersecurity measures can reduce duplication and enhance resilience.
6.4 Engage with standards, notified bodies and regulators
Given the evolving state of AI-relevant standards and regulatory interpretation, manufacturers benefit from proactive engagement with standard-setting bodies, notified bodies and national regulators. Participation in standardisation efforts allows firms to anticipate emerging expectations and to influence how AI Act requirements are concretely applied in industrial contexts.
Early dialogue with notified bodies can clarify how novel AI safety functions will be evaluated and what evidence will be expected in conformity assessments. Engagement with market-surveillance authorities can also help manufacturers understand reporting expectations and coordinate responses to incidents in ways that minimise disruption while meeting legal obligations.
Conclusion
The EU AI Act embeds AI firmly within the regulatory architecture that governs manufacturing products and processes in Europe. For manufacturing companies, AI components and systems are now subject to structured risk management, conformity assessment, documentation and market-surveillance obligations that parallel those for mechanical and electrical safety.
In practical terms, this means treating AI-based safety functions, industrial control systems, worker-management tools and quality-control applications as regulated elements of the industrial stack. The Act will influence product architectures, factory automation strategies, supplier relationships and global market access. Manufacturers that integrate AI governance into their core engineering and quality-management functions, invest in data and traceability infrastructure, and engage with evolving standards and regulatory practice will be better positioned to navigate this transition and to use compliant AI capabilities as part of a credible and resilient industrial strategy.
References
- European Commission, “AI Act – Shaping Europe’s digital future” (Digital Strategy portal).
- European Parliament, “EU AI Act: first regulation on artificial intelligence.”
- ArtificialIntelligenceAct.eu, “Article 3: Definitions” and “Article 6: Classification Rules for High-Risk AI Systems.”
- DIGITALEUROPE, “Use cases: How AI Act compliance obligations affect industrial AI deployment.”
- AI Value Institute, “AI Act for Industrial Systems: High-Risk Classification on the Plant Floor.”
- European research institutes, analyses on the interaction between the AI Act and the Machinery Regulation and other product-safety frameworks.
- Inrow.farm, “EU AI Act for agricultural machinery — what manufacturers and dealers need to know.”
- MMOWW / industry briefing, “EU AI Act Impact on Manufacturing: What Changes in August 2026.”
- Controls, Drives & Automation, “What machine builders need to know about the EU AI Act.”
- PA Consulting, “How the EU AI Act impacts medical device manufacturers.”
- KPMG, “Decoding the EU AI Act.”
- AzzurroDigitale, “AI Act: a guide for a safe factory towards August 2026.”
- Emerald Journal of Manufacturing Technology Management, “EU AI Act regulation: a study of non-European Union manufacturers.”
- Cambridge Journal of International and Comparative Law, “Introductory note to Regulation 2024/1689 of the European Parliament and Council.”
- 28Digital / EIT Digital, “Generative AI: Europe’s quest for regulation, innovation and competitiveness.”