本文目前仅提供英文版本。
AI and Robotics: Governance, Safety, and Liability When AI Takes Physical Form
When AI controls physical systems — industrial robots, surgical robots, autonomous drones, warehouse automation, delivery robots — governance moves beyond data and algorithms into physical safety, product liability, and human-robot interaction. The regulatory frameworks, liability questions, and governance requirements are fundamentally different from software-only AI.
Key Takeaways
AI-powered robots that interact with people or operate in shared physical spaces create safety, liability, and governance obligations that go beyond software AI governance.
The EU AI Safety regulation (part of the revised Machinery Regulation 2023/1230, effective January 2027) establishes specific requirements for AI in safety-critical machinery.
Product liability for AI-enabled robots is evolving — the EU's revised Product Liability Directive (2024/2853) explicitly covers AI and digital products, effective December 2026.
The Five Eyes agentic AI guidance (May 2026) directly applies to robotic systems that plan and execute actions autonomously.
Industrial robot safety standards (ISO 10218, ISO/TS 15066 for collaborative robots) must be integrated with AI governance requirements.
"仅供参考。本文不构成法律、监管、财务或专业建议。如需具体指导，请咨询合格专家。"
AI robotics governance is the discipline of managing the safety, liability, ethical, and regulatory risks created when artificial intelligence controls physical systems that interact with people and the physical world. This includes industrial manufacturing robots, surgical robots, warehouse and logistics automation, agricultural robots, autonomous drones, delivery robots, service robots in hospitality and retail, and collaborative robots (cobots) that work alongside humans. The governance challenge is fundamentally different from software-only AI: when an AI system controlling a physical robot makes an error, the consequences are physical — injury, property damage, environmental harm, or death. This creates a governance intersection between AI regulation, product safety law, workplace safety law, and product liability that most AI governance frameworks have not been designed to address.
Regulatory landscape for AI in robotics
The EU Machinery Regulation (2023/1230), which replaces the Machinery Directive and takes effect in January 2027, introduces specific requirements for AI-enabled safety components in machinery. Manufacturers must conduct risk assessments that account for AI-specific failure modes including unexpected behaviour, learning-based drift, and adversarial inputs. The EU AI Act classifies AI systems used as safety components of products covered by EU product safety legislation as high-risk under Article 6(1) and Annex I — with compliance now required by August 2028 following the Digital Omnibus extension.
The EU's revised Product Liability Directive (2024/2853), which member states must implement by December 2026, explicitly covers AI and digital products. It introduces a presumption of defectiveness when a manufacturer fails to disclose relevant information about their AI system, and extends liability to software providers — not just hardware manufacturers. This is a significant change for robotics companies that integrate third-party AI.
ISO standards for robot safety — ISO 10218 (industrial robots), ISO/TS 15066 (collaborative robots), and ISO 13482 (personal care robots) — establish physical safety requirements that must be integrated with AI governance. These standards address mechanical safety, force limiting, speed and separation monitoring, and human-robot interaction — but they were written before AI-driven autonomy was common and do not fully address AI-specific risks like model drift, unexpected learned behaviours, or adversarial manipulation.
Key governance requirements
Organisations deploying AI-powered robots should implement safety case governance — a documented, evidence-based argument that the system is acceptably safe for its intended context, maintained and updated throughout the system's operational life. This includes AI-specific validation testing (testing not just mechanical safety but AI decision-making under edge cases, adversarial conditions, and degraded inputs), human oversight mechanisms (clear procedures for human intervention, emergency stop, and manual override), continuous monitoring of AI system performance in production (detecting drift, unexpected behaviours, and degradation), incident response procedures specific to AI-robot failures (distinct from IT incident response), and supply chain governance for AI components (ensuring third-party AI models embedded in robotic systems meet your safety and governance requirements).
Further reading: EU AI Act | ISO 10218 — Industrial Robot Safety | Five Eyes Agentic AI Guidance