
ISO/IEC 42001 sistem upravljanja veštačkom inteligencijom, Foundation (ICO AIMS FND)
Dvodnevni uvod u sisteme upravljanja veštačkom inteligencijom prema ISO/IEC 42001. Naučite šta standard zahteva, kako funkcionišu kontrole iz Aneksa A i gde se sa njima susreće EU AI Act. Priprema vas za ispit ICO AIMS 42001:2023 FND. Prethodno iskustvo sa standardima nije potrebno.
Sledeći termini
- 7decLive online (Zoom)
- 4janLive online (Zoom)
- 1febLive online (Zoom)
O ovom treningu
Od svake organizacije koja razvija ili kupuje veštačku inteligenciju danas se očekuje da pokaže da tu tehnologiju drži pod kontrolom. EU AI Act je od toga napravio pravno pitanje, a ne samo dobru praksu. ISO/IEC 42001 je odgovor na kome se tržište ustalilo: prvi međunarodni standard za sistem upravljanja veštačkom inteligencijom, okvir kojim kompanija dokazuje da se njenom AI upravlja, a ne improvizuje.
Ovaj Foundation trening vas tokom dva dana sa predavačem vodi kroz taj standard. Učite rečnik, strukturu sistema upravljanja, kontrole iz Aneksa A i smernice iz Aneksa B, i dovoljno o samoj veštačkoj inteligenciji da sve to primenite. Posle treninga umete da čitate standard, pratite projekat sertifikacije i razgovarate sa auditorima ravnopravno.
Kome je trening namenjen
- Menadžerima kvaliteta, usklađenosti i rizika čija organizacija uvodi AI
- IT i data rukovodiocima od kojih se traži da AI projekte uvedu u sistem upravljanja
- Konsultantima i projekt menadžerima koji pripremaju ISO/IEC 42001 sertifikaciju
- Svima koji već poznaju ISO 9001 ili ISO/IEC 27001, a sada im treba AI pandan
- Budućim AIMS Officer-ima i AIMS auditorima, kojima je ovo prvi korak
Šta ćete naučiti
- Da precizno koristite pojmove ISO/IEC 42001: sistem upravljanja, politika, cilj, rizik, kontrola, usaglašenost, procena uticaja
- Da objasnite kako se gradi sistem upravljanja AI, od konteksta organizacije preko planiranja i rada do stalnog poboljšanja
- Da radite sa Aneksom A: šta pokrivaju ciljevi kontrola i kako se sastavlja Izjava o primenljivosti
- Da primenite smernice iz Aneksa B za izbor i obrazloženje kontrola za stvaran slučaj upotrebe AI
- Da opišete kako se AI sistemi grade i treniraju, dovoljno da procenite gde leže rizici
- Da postavite EU AI Act, GDPR i srodne obaveze uz standard i vidite gde se šta primenjuje
Program treninga
Prvi dan, sistem upravljanja
- Kako su struktuirani trening i ispit
- Pojmovi i definicije, zajednički rečnik na kome standard počiva
- Sistem upravljanja AI u celini: kontekst, liderstvo, planiranje, podrška, rad, vrednovanje, poboljšanje
Drugi dan, kontrole, AI i pravo
- Aneks A, normativni ciljevi kontrola i kontrole
- Aneks B, smernice za primenu AI kontrola
- Uvod u veštačku inteligenciju: kako sistemi kojima upravljate zaista rade
- Zakoni o AI i relevantni propisi, sa EU AI Act-om u središtu
- Priprema za ispit i primeri pitanja
Preduslovi
Nema. Trening je zamišljen kao ulazna tačka. Iskustvo sa bilo kojim ISO sistemom upravljanja pomaže, ali nije obavezno; tehničko ili programersko predznanje se ne pretpostavlja.
Ispit i sertifikacija
Trening vas priprema za ispit ICO AIMS 42001:2023 FND, koji dodeljuje ICO International Certification Organization GmbH. To je akreditovan sertifikacioni ispit, a ne potvrda o pohađanju.
- 30 pitanja sa višestrukim izborom, gde jedna, više ili sve opcije mogu biti tačne
- 45 minuta, bez literature
- Prag prolaznosti 60 procenata
- Polaže se onlajn, u vreme i na mestu po vašem izboru
Ispit je dostupan kao opciona dopuna treninga. Sve detalje objavljuje telo koje dodeljuje sertifikat na ico-cert.org.
Šta je uključeno
- Dva dana, 12 sati, obuke sa predavačem na engleskom, onlajn ili uživo
- Male grupe, nikad više od 28 polaznika
- Kompletni materijali, vežbe i primeri ispitnih pitanja
- Razrađeni primeri iz evropskih organizacija u svakom bloku
- Potvrda o pohađanju, uz ispit ICO AIMS 42001:2023 FND kao opcionu dopunu
Program kursa
Poglavlja prate zvanični ICO syllabus. Procenat je udeo poglavlja u ispitu.
The syllabus of this course is divided into 4 key areas:
- Introduction and terms
- Artificial Intelligence Management System with Annex A and Annex B
- Introduction to Artificial Intelligence
- AI laws and relevant regulations
Term Definitions
- Management System
A system of interconnected elements within an organization designed to establish policies, set objectives, and define processes to achieve those objectives.
- Policy
The formally expressed intentions and direction of an organization as determined by its top management.
- Objective
A desired result or goal to be achieved.
- Risk
The impact of uncertainty on achieving objectives.
- Process
A sequence of related or interacting activities that use inputs to deliver an intended result.
- Documented Information
Information that an organization must control and maintain, regardless of its medium.
- Performance
A measurable result achieved through processes or activities.
- Continual Improvement
Ongoing efforts to enhance processes and performance.
- Effectiveness
The degree to which planned activities are executed and intended results are achieved.
- Requirement
A need or expectation that is explicitly stated, generally implied, or mandatory.
- Conformity
The fulfillment of a specified requirement.
- Nonconformity
The failure to meet a specified requirement.
- Corrective Action
An action taken to eliminate the root cause(s) of a nonconformity and prevent its recurrence.
- Audit
A systematic, independent process to gather evidence and objectively evaluate whether specified criteria are met.
- Control
A measure that maintains or modifies a risk.
- Information Security
The protection of confidentiality, integrity, and availability of information.
- AI System Impact Assessment
A formal, documented process for identifying, evaluating, and addressing impacts on individuals, groups, or society from AI systems.
- Statement of Applicability
A documented overview of all necessary controls and justification for their inclusion or exclusion.
- Confidentiality
One of the three main aspects of IT security. It ensures that information is accessible only to authorized individuals and systems, and is not disclosed without permission.
- Integrity
One of the three main aspects of IT security. It guarantees that data and systems remain accurate, unaltered, and reliable, unless changes are made by authorized individuals or systems.
- Availability
One of the three main aspects of IT security. It ensures that information and systems are accessible in a timely and reliable manner to authorized individuals and systems.
- PDCA Lifecycle
A process model based on the ISO standard high-level structure. The cycle includes the phases Plan, Do, Check, and Act, ensuring continuous improvement.
Resources
Management system in general
A management system is a structured framework that supports organizations in systematically achieving specific objectives. It comprises guidelines, procedures, processes, responsibilities and resources. The aim is to ensure the effectiveness and efficiency of the organization in a specific area - such as quality, information security or here: Artificial Intelligence - and to continuously improve it.
Management systems are characterized by the PDCA cycle (Plan-Do-Check-Act):
- Plan: define objectives, resources and processes.
- Do: Implementation of the planned measures.
- Check: Review the results (e.g. through internal audits).
- Act: Correction and improvement.
Summary of the Seven ISO 9001 Quality Management Principles
- Customer Orientation
Identifying and meeting customer requirements to build trust and ensure long-term satisfaction. - Leadership
Establishing a clear vision and direction, motivating leaders, and empowering employees to actively shape quality objectives. - Engagement of People
Leveraging the competence, motivation, and knowledge of all employees through participation to foster ownership and gather improvement ideas. - Process-Oriented Approach
Managing activities and resources as interconnected processes to enhance consistency, efficiency, and transparency throughout the value chain. - Improvement
Systematically optimizing the organization, its products, and processes through goal setting, action planning, and regular review. - Evidence-Based Decision Making
Basing decisions on reliable data and facts, conducting analyses, and monitoring key performance indicators. - Relationship Management
Maintaining partnerships with suppliers and stakeholders to ensure sustainable supply and mutual benefit.
Artificial Intelligence Management System (AIMS)
An Artificial Intelligence Management System (AIMS) in accordance with ISO/IEC 42001 is a management system specifically for the use and control of artificial intelligence (AI). It supports organizations in identifying and managing the opportunities and risks of AI systems and ensuring the responsible use of AI.
Key aspects of an AIMS are
- Responsibility and leadership: defining leadership roles, responsibilities and communication channels.
- Risk-based approach: Identification and assessment of risks associated with AI systems.
- Documentation and control: Creation and maintenance of relevant documents, e.g. guidelines, procedural instructions and records.
- Internal audits: Regular reviews by the organization itself to ensure compliance and effectiveness of the system.
- Management review: Regular assessment by top management to ensure the adequacy and effectiveness of the system.
- Continuous improvement: Identification of weaknesses and derivation of improvement measures.
Context of the organization (ISO 42001 Chapter 4)
- Understand the organization and its context: Analyze internal and external factors that affect the organization's ability to achieve its intended outcomes
- Understand the needs and expectations of interested parties: Identify relevant stakeholders and their requirements.
- Define the scope of the AI management system: Define the boundaries and scope of the system.
- AI management system: The organization must establish, implement, maintain, continuously improve and document an AI management system, including necessary processes and their interactions.
Leadership (ISO 42001 Chapter 5)
- Leadership and engagement
- KI policy
- Roles, responsibilities and authorities
Planning (ISO 42001 Chapter 6)
- Measures to manage risks and opportunities
- General
- KI risk analysis
- KI risk treatment
- Evaluation of the impact of the AI system
- AI goals and planning to achieve them
- Planning for changes
Support (ISO 42001 Chapter 7)
- Resources
- Competence
- Awareness
- Communication
- Documented information
Operations (ISO 42001 Chapter 8)
- Operational planning and control
- KI risk analysis
- KI risk treatment
- Evaluation of the impact of the AI system
Performance evaluation (ISO 42001 Chapter 9)
- Monitoring, measurement, analysis and evaluation
- Internal audits
- Management review
Improvement (ISO 42001 Chapter 10)
- Continuous improvement
- Nonconformities and corrective actions
Sources
Annex A ISO/IEC 42001:2023 (normative) – Reference Control Objectives and Controls
1 General
The controls outlined in Table A.1 offer organisations guidance for achieving their business objectives and managing the risks that arise when designing and operating AI systems. The entire set of control objectives and controls in Table A.1 does not have to be adopted; organisations may define and implement their own controls (see 6.1.3). Annex B supplies implementation guidance for every control listed in Table A.1.
2 AI-Related Policies
Objective: To establish management direction and support for AI systems that aligns with the organisation’s business requirements.
3 Internal Organisation
Objective: To allocate clear accountability within the organisation so that a responsible approach to the implementation, operation and governance of AI systems is maintained.
4 Resources for AI Systems
Objective: To ensure the organisation accounts for every resource— including AI system components and assets—so that risks and impacts can be fully identified and addressed.
5 Assessing Impacts of AI Systems
Objective: To evaluate the effects that an AI system may have on individuals, groups and society throughout its entire life cycle.
6 AI System Life Cycle
6.1 Management Guidance for AI System Development – Objective: To identify and document goals and to implement processes that foster the responsible design and development of AI systems.
6.2 AI System Life-Cycle Management – Objective: To define criteria and requirements for each stage of the AI system life cycle.
7 Data for AI Systems
Objective: To ensure that the organisation understands the role and impact of data during the development, deployment and use of AI systems across their life cycles.
8 Information for Interested Parties
Objective: To provide relevant stakeholders with the information they need to understand and assess both the positive and negative risks and impacts.
9 Use of AI Systems
Objective: To ensure that AI systems are used responsibly and in accordance with organisational policies.
10 Third-Party and Customer Relationships
Objective: To confirm that the organisation recognises its responsibilities, remains accountable and distributes risks appropriately whenever third parties are involved at any stage of the AI system life cycle.
1 Fundamentals
The Annex B of ISO 42001 serves as a starting point for developing organization-specific control mechanisms.
The documented implementation guidelines refer to the controls listed in Table A.1. They provide supporting information for control implementation without requiring organizations to justify the inclusion or exclusion in the applicability statement.
The implementation guide is not universally applicable and may not cover all organization-specific control needs. Organizations can expand the guide according to their individual risk management requirements, modify it, or develop their own control strategies.
The annex serves as a guidance framework for establishing and implementing controls to address AI risks within the defined AI management system.
Chapters and Objectives
- AI Policies: Implementation of management instructions and support for AI systems in accordance with business requirements
- Internal Organization: Accountability within the organization for a responsible approach to AI systems
- Resources for AI Systems: Ensuring comprehensive understanding of AI system resources and their risks
- AI System Impact Assessment: Analysis of impacts on individuals, groups, and societies
- AI System Lifecycle: Identification and documentation of goals for responsible development
- Data for AI Systems: Understanding the role and impacts of data in the AI system lifecycle
- Information for Stakeholders: Provision of necessary information for risk assessment
- AI System Deployment: Responsible implementation in accordance with corporate guidelines
- Relationships with Third Parties and Customers: Ensuring responsibilities and appropriate risk distribution
General
Artificial Intelligence (AI) refers to the ability of machines to perform tasks that typically require human intelligence. This includes learning from data, recognizing patterns, and making decisions. AI is used to develop systems such as language assistants, image recognition, and translation services.
Basic Principles of AI
Model Training: AI models are trained with large datasets to recognize patterns and make decisions based on them.
Quality of Training Data: The accuracy of results directly depends on the quality of the data used for training. Poor or incorrect training data leads to inaccurate and faulty outcomes.
Conclusion: The power of an AI system lies in the quality of the data it is based on. To achieve reliable results, training data must be precise and representative.
Introduction to Generative AI
Generative artificial intelligence (known as generative AI or GenAI) describes a specialized area of artificial intelligence aimed at generating new data based on existing datasets. Generative models, particularly neural networks like GANs (Generative Adversarial Networks) and Transformer architectures, analyze and learn the underlying patterns of their training data to create synthetic content that can appear both creative and realistic.
The technological breakthrough in the early 2020s is due to advances in transformer-based neural networks. Large language models (LLMs) such as GPT (Generative Pre-trained Transformer) and systems built on it like ChatGPT have taken generative AI to the next level. These models can not only carry out fluid dialogues but also perform highly complex tasks such as writing code, creating scientific reports, synthesizing multimodal content (text, image, video), or providing advanced Text-to-Speech (TTS) and Speech-to-Text (STT) capabilities.
Notable applications include:
- Chatbots: ChatGPT, Gemini, and Copilot, which enable human-like conversations.
- Document Generation: Automated creation of documents integrating text, images, audio, etc.
- Image Generation: Stable Diffusion, Midjourney – generating visuals from text prompts.
- Video Generation: e.g., Sora, which generates video from textual input.
- Text-to-Speech (TTS) / Speech-to-Text (STT): natural language audio interaction.
Multimodality: The ability of AI to process and combine text, image, audio, and video.
Major companies like OpenAI, Microsoft, and Google are leading development and commercialization of GenAI.
Special Features of Generative AI
- Creativity: Artistic/musical/written content creation based on training data.
- Adaptability: Can respond to diverse inputs and contexts.
- Pattern Recognition: Understands and recreates complex patterns.
- Handling Complexity: Capable of managing and synthesizing large data sets.
- Text & Speech Integration: Seamless real-time communication with users.
Applications of Generative AI
Content creation, design, accessibility, education, marketing, healthcare, and virtual assistants.
What AI Cannot or Does Poorly
- Feelings & Emotions
- Ethical/Moral Decision-Making
- Common Sense
- Self-Awareness
- Contextual Understanding
- Incomplete/Faulty Training Data
- Intuition
- Cultural & Social Nuances
Risks of Using AI
Note: Non-exhaustive list of examples by stakeholder perspective.
Business Perspective
- Legal Risks: GDPR violations, EU AI Act non-compliance, liability concerns.
- Technical Risks: Bias, system integration issues, hallucinations.
- Reputational Risks: Loss of trust, negative media.
Employee Perspective
- Job insecurity, fear of change, lack of skills
- Misuse, overreliance, poor interpretation
Customer Perspective
- Data privacy breaches
- Wrong decisions, unfair treatment
- Lack of transparency, manipulation
Other Stakeholders
- Ethical/Social impact
- Over-automation, inequality reinforcement
- Platform dependency
- Cybersecurity threats
Chances in Using AI
- Automation: AI enables the automation of repetitive and time-consuming tasks, increasing overall operational efficiency.
- Decision Support: AI can provide accurate, data-driven insights to support strategic and operational decision-making.
- Personalization: AI systems allow for tailored experiences in customer service, learning environments, and product recommendations.
- Innovation: AI opens up new possibilities in product development, service design, and scientific research.
- Predictive Capabilities: Early identification of risks, trends, and opportunities through pattern recognition in large data sets.
- Enhanced Accessibility: Tools such as text-to-speech, speech-to-text, and real-time translation assist people with disabilities and support multilingual interaction.
- Resource Optimization: AI can help reduce energy consumption, optimize logistics, and manage infrastructure proactively.
- 24/7 Availability: AI-powered systems (e.g., chatbots) can operate continuously without human intervention.
- Competitive Advantage: Organizations using AI effectively can gain a market edge through faster innovation cycles and better customer understanding.
Summary of the EU AI Act
The EU AI Act is a comprehensive regulatory framework of the European Union for Artificial Intelligence (AI). The aim is to make the development and use of AI safe, transparent, and ethical. It is based on a risk-based approach and contains the following key points:
Classification of AI systems according to risk levels
- Prohibited AI: Systems with unacceptable risks (e.g. systematic evaluation of socially unacceptable behavior, manipulative practices).
- High-risk AI: Areas of application with high potential impact, e.g.:
- Medical diagnostics
- Critical infrastructures
- Evaluation of work performance
- Limited-risk AI: AI systems with limited risk, but which must meet transparency requirements (e.g. chatbots).
- Minimal risk AI: AI systems with no or low risk (e.g. AI-powered games).
Central requirements for AI
- Transparency: Users must be informed about how it works and how it is used.
- Data management: Ensuring quality, accuracy and fairness of data.
- Monitoring: People must be able to maintain control and monitor AI systems.
- Compliance assessment: Manufacturers must ensure that systems comply with EU regulations.
Aim of the law
To promote innovation while minimizing risks to citizens' rights and security.
Article 4 of the EU AI Act
Article 4 of the EU AI Act, as amended by Regulation (EU) 2026/1744, asks providers and deployers of AI systems to take measures that support the AI literacy of their staff and of other persons who operate or use AI systems on their behalf, taking into account their role, their prior knowledge and the context in which the systems are used. It does not define a required level and it does not ask for a formal proof; it does expect organisations to be able to show which measures they took. Training with a defined syllabus and an externally set exam is one such measure.
The exact wording is in the consolidated text of Regulation (EU) 2024/1689 on EUR-Lex (see Sources).
Summary of Regulated AI in the EU AI Act
Regulated AI includes all systems that are either prohibited, classified as high risk, or limited risk and thus subject to the requirements of the Act. The aim is to minimize risks for citizens, companies, and institutions while promoting innovation in AI.
Other relevant laws in the EU
1. General Data Protection Regulation (GDPR)
- Regulates the handling of personal data.
- Relevant for AI that processes personal data:
- Transparency obligations
- Rights of data subjects: (e.g. information, erasure, automated decisions)
- Data protection impact assessment for high-risk systems.
2. Product liability law
- Addition to the AI Act: manufacturers can be held liable if their AI systems cause damage.
- Planned adaptation of the Product Liability Directive to include AI.
3. Labor law
- Use of AI for monitoring or making decisions about employees (e.g. hiring) must be legally sound.
4. Copyright
- Clarification of property rights for AI-generated content in accordance with copyright law.
Sources
Akreditacije i sertifikacije
Ovaj trening vas priprema za sledeće sertifikacione šeme.
Predstojeći događaji
Ili pozovite telefonom:011 635 5918
