From 20 January 2027, the Machinery Regulation (EU) 2023/1230 replaces the Machinery Directive 2006/42/EC across the EU. For robotics and automation companies the most important change is that software is now explicitly within scope as a safety component, cybersecurity threats to safety functions must be addressed, and machine learning that ensures safety functions triggers mandatory third-party conformity assessment. If you use an external engineering team on robot or machine software, these changes determine where the boundary of their work should be. The short answer: an external team can contribute substantially to non-safety software, test infrastructure and documentation, while safety functions should stay inside a process that you, as manufacturer, control end to end.

The industry context

The new regulation lands on an industry under strain. The VDMA reported a real production decline of 5 percent for German mechanical engineering in 2025 and forecast growth of only 1 percent for 2026, well below the 2023 peak of 255 billion euros. The sector still employed slightly more than 1 million people in Germany in September 2025, including 190,000 engineers, but according to the VDMA around 100,000 jobs have been lost since 2019. Automation intensity keeps rising regardless: the IFR World Robotics 2025 report puts Germany's robot density at 449 industrial robots per 10,000 manufacturing employees, fourth worldwide, with 26,982 new installations, the highest in Europe. Software is moving deeper into the machine as well. Bitkom's 2026 Industrie 4.0 study found that 40 percent of German manufacturers with 100 or more employees already use AI in production and a further 38 percent plan to. Machine builders therefore face more safety-relevant software and more conformity work with fewer engineers, and the Cyber Resilience Act becomes fully applicable on 11 December 2027, less than a year after the Machinery Regulation.

What changes in 2027

  • Software as a safety component. The definition of safety component now covers software, including software placed on the market separately. A safety-relevant software update is no longer an informal patch; it is a change to a safety component.
  • Self-evolving behaviour. Annex I Part A lists, among the categories that require third-party conformity assessment, safety components with fully or partially self-evolving behaviour using machine learning approaches that ensure safety functions, and machinery with embedded systems of that kind. For these categories, applying harmonised standards does not allow self-certification.
  • Protection against corruption. The essential health and safety requirements include protection against corruption, meaning that connections and software must be designed so that malicious or accidental interference cannot lead to hazardous situations. EN 50742 is being positioned as the relevant standard for this requirement.
  • Traceability of safety software. The requirements on control systems expect evidence of interventions and of safety software versions loaded after the machine is placed on the market.
  • Substantial modification. Whoever substantially modifies machinery in a way that affects safety is treated as the manufacturer for that modification. This matters for integrators and for operators who add autonomy features to existing machines.

Machinery is also listed in Annex I of the AI Act. For AI systems that are safety components of machinery, the high-risk AI obligations apply from 2 August 2028 after the Digital Omnibus, which gives some time but adds a second layer of documentation.

The standards stack

For robotics, the relevant harmonised and state-of-the-art standards include ISO 12100 (risk assessment), ISO 13849-1 and IEC 62061 (functional safety of control systems), and the 2025 revision of ISO 10218-1 and ISO 10218-2 for industrial robots and robot systems, which incorporates requirements for collaborative applications. Mobile robots in industrial settings typically also draw on ISO 3691-4. The specific selection depends on your machine and should come from your safety engineer, not from a vendor.

A decision framework: four zones

The most useful way to allocate work is to classify each software item into one of four zones, based on its relationship to safety functions.

ZoneExamplesWho should own it
A. Safety functionsSafety PLC logic, safe torque off, speed and separation monitoring, safety-rated zonesManufacturer's safety team under ISO 13849 or IEC 62061 process. External input only as reviewed contributions
B. Safety-adjacent softwareConfiguration tools for safety zones, diagnostics reading safety state, interfaces that could influence safety parametersShared. External team may build, manufacturer verifies and controls release
C. Non-safety functional softwareNavigation, task planning, perception for productivity, HMI, fleet managementExternal team can own, with architectural proof that zone C cannot override zone A
D. Engineering infrastructureSimulation, test automation, CI, traceability tooling, documentation generationExternal team can own fully

The critical design decision is the separation between zone A and zone C. If the autonomy stack can only request motion and the safety layer independently enforces limits, a defect in navigation software produces a stop, not a collision. That separation is also what allows non-safety software to iterate quickly without each change becoming a safety change.

Where machine learning complicates the picture

Perception models increasingly contribute to safety concepts, for example person detection used to reduce speed. If such a model ensures a safety function and has self-evolving behaviour, the Annex I Part A category applies. Many teams therefore deliberately keep learned components in zone C (productivity) and ensure safety through certified sensors and conventional logic in zone A. If you choose to let a learned model carry a safety function, plan for notified body involvement and for AI Act obligations from August 2028, and do not delegate that design decision to an external team.

Cybersecurity is now a safety topic

Protection against corruption links the Machinery Regulation to industrial security practice. Remote engineering teams change your attack surface: build pipelines, signing keys and remote access paths become relevant to safety. Minimum controls:

  • Signed software artefacts, with signing keys held by the manufacturer, not by the external team.
  • No direct remote access from external networks to machines in the field. Access through a manufacturer-controlled jump host, with session logging.
  • Software bill of materials for every release, so that vulnerable components can be identified quickly.
  • Separation of build environments for zone A software.

Contract wording that reflects the zones

The zone model only works if it appears in the contract. In the Leistungsbeschreibung, name the zones the external team works in, list the repositories and components that belong to zone A, and state that changes to those components are delivered as proposals for review rather than as releases. Acceptance of zone B deliverables should require verification by your safety engineer. This keeps release authority where the Machinery Regulation places responsibility: with the manufacturer.

Checklist for 2026 and 2027

  1. Inventory all software on your machines and classify each item into zones A to D.
  2. Verify that zone C software cannot bypass zone A limits, and document the argument.
  3. Check whether any machine learning component ensures a safety function.
  4. Define a release process for safety software that records versions and interventions.
  5. Review your risk assessment for cyber threats to safety functions.
  6. Specify in contracts with external teams which zones they work in and who approves releases in zones A and B.
  7. Keep signing keys and production access under your control.

This article is general information and not legal advice. The applicable requirements depend on your specific machine; consult your safety engineer, notified body or a qualified lawyer.

The Machinery Regulation turns the boundary between safety functions and everything around them into a question of evidence, not organisation. niogod™ pods are scoped for zones C and D, with clearly bounded contributions to zone B under your review, delivered under a Werkvertrag with Abnahme per sprint and coordinated in German by the Projektleiter with your safety engineers. If 20 January 2027 is on your compliance plan, the scoping conversation is worth having now.

Talk to niogod™

Planning engineering capacity in DACH? Book a 30-minute consultation with the niogod™ team.

Book consultation
← All insights