Overview - ISO/IEC TS 8200:2024 (Controllability of AI systems)
ISO/IEC TS 8200:2024 is a Technical Specification from ISO/IEC JTC 1/SC 42 that defines a basic framework, principles and approaches to achieve and enhance controllability of automated AI systems. Applicable across the AI life cycle and to all types of organizations (commercial, government, not‑for‑profit), the document focuses on how controllers (humans or other external agents) can observe, influence and transfer control of AI systems to ensure predictable, safe behaviour.
Key topics and technical requirements
The specification covers practical, testable aspects of controllability rather than theoretical concepts. Core topics include:
- State observability and state transition - defining what system state means, how it changes, and observable indicators for control decisions.
- Control transfer process and cost - where and how control is transferred, and considerations for time, complexity, energy and other costs of transfer.
- Reaction to uncertainty during control transfer - identifying and mitigating uncertainty on both the AI system and controller sides.
- Verification and validation approaches - recommended verification processes, validation (including retrospective validation), and expected verification/validation outputs.
- Characteristics of controllability - control points, span of control, engagement/disengagement, collaborative control and controllability levels (including continuous learning systems).
- Design and implementation guidance - principles and lifecycle stage guidance (inception, design, training data, architecture, risk management, safety‑critical considerations).
- Testing focus - both functional and non‑functional testing for controllability and example documentation in informative annexes.
The document defines requirements on controllability and suggests methods for cost estimation, handling side effects and integrating controllability into AI system architecture and development processes.
Practical applications and who should use it
ISO/IEC TS 8200:2024 is directly useful for:
- AI system designers and software architects implementing control points and safe state transitions.
- Safety engineers and risk managers integrating controllability into hazard analysis and mitigation.
- DevOps, MLOps and QA teams responsible for verification, validation and testing of AI behaviour.
- Regulators, procurement officers and auditors specifying controllability expectations for AI deployments.
- Organizations operating continuous learning or adaptive AI systems that require ongoing controllability assurance.
Use cases include safety‑critical systems, automated decision support, autonomous operations, and any deployment where human or external oversight must reliably regain or share control.
Related standards
- ISO/IEC 22989:2022 (automated systems and AI lifecycle concepts) - referenced as contextual foundation.
- Other ISO/IEC AI governance and lifecycle standards from JTC 1/SC 42.
Keywords: ISO/IEC TS 8200:2024, controllability, AI systems, control transfer, state observability, verification and validation, continuous learning, AI lifecycle.