ISO/IEC 27091:202x (DIS) - Cybersecurity and Privacy for Artificial Intelligence

0.1 General

This document provides guidance for organizations to address privacy risks in artificial intelligence (AI) systems and machine learning (ML) models. It extends ISO/IEC 27001 principles to AI-specific privacy concerns.

It applies to organizations developing or using AI systems that process personal data.

Example: A company using AI for customer profiling must implement privacy risk controls to avoid data breaches or unlawful processing.

0.2 Relationship with Other Standards

This standard complements ISO/IEC 27001 (ISMS) and ISO/IEC 42001 (AI Management System), focusing on privacy in AI.

Example: Use with ISO/IEC 27001 for overall cybersecurity and add AI-specific privacy measures from this standard.

This standard offers guidance on managing privacy risks in AI and ML systems, including identification, assessment, and mitigation.

Example: For an AI chatbot handling user data, assess risks like data leakage and mitigate with encryption.

Key references include:

- ISO/IEC 27001: Information security management systems

- ISO/IEC 42001: Artificial intelligence � Management system

- EU AI Act (for context)

Example: Align privacy controls with ISO/IEC 27001 Annex A for AI-specific applications.

Key terms:

- Privacy Risk: Risk of adverse effects on individuals' privacy from AI processing.

- AI System: Machine-based system with autonomy that infers from inputs to generate outputs.

- Machine Learning Model: AI component that learns from data.

Example: "Privacy Impact" refers to how AI inference might reveal sensitive information unintentionally.

4.1 Understanding the Organisation and its Context

Identify external and internal issues affecting privacy in AI, like regulations and data sources.

Example: Consider GDPR for EU users and internal data policies.

4.2 Understanding the Needs and Expectations of Interested Parties

Determine stakeholders' privacy expectations, such as users and regulators.

Example: Users expect data anonymity; regulators require compliance reports.

4.3 Determining the Scope of the Privacy Management System

Define boundaries for privacy management in AI.

Example: Scope includes all AI models processing personal data but excludes non-AI systems.

5.1 Leadership and Commitment

Top management must commit to privacy in AI, providing resources and direction.

Example: CEO mandates privacy-by-design in all AI projects.

5.2 Policy

Establish a privacy policy for AI, committing to compliance and risk management.

Example: Policy includes "All AI data processing will minimize privacy risks."

5.3 Organizational Roles, Responsibilities and Authorities

Assign roles for privacy management in AI.

Example: DPO oversees AI privacy compliance.

6.1 Actions to Address Risks and Opportunities

Plan actions to handle privacy risks in AI.

Example: Identify risk of data inference in ML and plan mitigation like differential privacy.

6.2 Privacy Objectives and Planning to Achieve Them

Set measurable privacy objectives for AI.

Example: Objective: "Reduce privacy breach incidents to zero annually."

7.1 Resources

Provide resources for privacy management in AI.

Example: Allocate budget for privacy-enhancing technologies.

7.2 Competence

Ensure staff competence for AI privacy tasks.

Example: Train developers on privacy risks in ML.

7.3 Awareness

Make staff aware of privacy policy and implications.

Example: Workshops on AI privacy risks.

7.4 Communication

Establish communication for privacy issues.

Example: Internal reporting channel for privacy concerns.

7.5 Documented Information

Maintain documented information for privacy management.

Example: Privacy impact assessments for AI models.

8.1 Operational Planning and Control

Plan operations to meet privacy requirements in AI.

Example: Integrate privacy checks in AI development pipeline.

8.2 Privacy by Design and by Default

Implement privacy from the start in AI systems.

Example: Use anonymization by default in data processing.

9.1 Monitoring, Measurement, Analysis and Evaluation

Monitor privacy performance in AI.

Example: Track data access logs for unauthorized use.

9.2 Internal Audit

Audit privacy management system.

Example: Annual internal audit of AI data handling.

9.3 Management Review

Review privacy management effectiveness.

Example: Quarterly management review of privacy incidents.

10.1 General

Improve privacy management system continually.

Example: Update privacy controls based on new threats.

10.2 Nonconformity and Corrective Action

Handle nonconformities and correct them.

Example: If privacy breach occurs, investigate and fix the cause.

10.3 Continual Improvement

Use evaluations to improve.

Example: Enhance privacy training after audit feedback.