INDUSTRY PERSPECTIVE / MANUFACTURING

Manufacturing AI Must Respect the Physical World

When digital intelligence can influence production, quality, and safety, integrity and availability become first-class AI requirements.

Manufacturers hold engineering drawings, process parameters, maintenance histories, quality records, supplier data, field knowledge, and operational-technology telemetry. Connecting these sources can reduce downtime and preserve expertise, but errors or unauthorized changes can affect equipment, product quality, safety, and delivery.

Separate knowledge assistance from control

An assistant that finds a maintenance procedure is not the same as a system that changes a setpoint. Define boundaries between read-only knowledge, recommendations, workflow actions, and direct OT control. Require stronger validation, authorization, and fail-safe behavior as the system approaches the physical process.

Preserve engineering provenance

Answers should identify drawing revision, asset, site, configuration, effective date, and approval state. Retired instructions and draft specifications must not appear as current truth.

Factory-ready controls
  • Segment AI services from safety-critical and control networks.
  • Use read-only integration unless action is explicitly engineered.
  • Protect intellectual property across suppliers and sites.
  • Test degraded, offline, adversarial, and emergency conditions.
  • Maintain manual procedures and recovery capability.

Optimize for resilience

Manufacturing environments often prioritize availability and integrity differently from conventional IT. AI should fit maintenance windows, legacy constraints, and incident recovery. The strongest first uses create value while leaving deterministic safety controls intact.

Reference: NIST Cybersecurity Framework Manufacturing Profile.