AI Governance and compliance.
AI digital workers operate within the organization’s policies. Check which requirements cover your use, which AiTrust control mechanisms relate to them and what the record confirms.
The scope of obligations follows from the organization’s role and the AI use. Deploying the tool supports specific technical activities; it does not replace legal and organizational assessment.
From requirement to evidence: how to use the map
- Step 1: Role and useDetermine what AI is used for in the process and what role the organization has in it.
- Step 2: RequirementsIdentify the laws, supervisory documents, standards and contracts that apply.
- Step 3: Control mechanismLink the requirement to a specific mechanism in the work path.
- Step 4: Evidence and its limitsCheck what the record confirms and what it does not cover.
- Step 5: Organization’s decisionsRecord what remains for legal and organizational assessment.
Requirement, mechanism, evidence, limit
A short summary of the most important topics. Details and the other topics are in the map below.
| Requirement | AiTrust mechanism | Evidence | Limit |
|---|---|---|---|
| AI Act: human oversight of high-risk systems (Art. 14) | Operations of specified classes require a human decision before execution. The approval threshold is set by the agent passport under which the AI digital worker operates. | The pre-send control panel with the user’s decision and the agent passport with the settings of the AI digital worker (real examples, synthetic data). | Whether a use is a high-risk system is assessed by the organization; the tool does not settle whether Art. 14 is met. |
| GDPR: minimizing the data passed to the model | The model receives the context after sanitization of specified data categories. | The version sent to AI with tags in place of the data and a list of redactions in the recorded run. | Applies to the specified categories and configuration; it does not replace the legal basis for processing, contracts or an impact assessment. |
| ISO/IEC 42001: AI management system in the organization | Rules, roles and records in the controlled path can be part of the management system. | Audit log view with filters and export (a separate example). | Certification applies to the organization; neither the tool nor its records are a certificate. |
| Audit and documenting work with AI | Evidence trail for operations in the controlled path; the organization sets the evidence retention period. | AI send report with the result and the audit log view. | The run record is not complete audit material or an impact assessment (DPIA). |
Requirements, mechanisms and evidence
Each topic shows the source of the requirement and its type, who it concerns, the related AiTrust mechanism, the available evidence with the limits of what it confirms, and what remains for the organization to decide.
AI Act: risk management and human oversight
EU lawSource of the requirement
Regulation (EU) 2024/1689 on artificial intelligence (AI Act), including the requirements for high-risk AI systems: Art. 9 (risk management system) and Art. 14 (human oversight).
Who it concerns
Obligations depend on the risk level and classification of the use and on the organization’s role, e.g. as the provider of an AI system or as an entity that deploys it. Oversight measures should match the risk, the level of autonomy and the context of use.
AiTrust mechanism
- Operations of specified classes require a human decision before execution.
- In the AiTrust application, the AI digital worker operates as an agent based on the agent passport. The passport records the maximum oversight zone, the approval threshold and the operation and cost limits, and the extent of their enforcement depends on the deployment. The passport can be suspended or revoked.
- Evidence trail for operations in the controlled path.
Evidence and the limits of what it confirms
The record of decisions and events can be part of the documentation the organization uses to meet the relevant requirements. It does not replace classifying the use, assessing the risk or determining who exercises oversight and with what competence.
For the organization to decide
- The organization’s role and the classification of the AI use.
- Who exercises oversight and what competence they have.
- Which operations require approval, and when the AI result should be disregarded or changed or the system stopped.
Next step
Determine the classification of the use and the points where a human decision is needed. In the demo, we will show how to map them in the work path.
GDPR: personal data and access in work with AI
EU lawSource of the requirement
Regulation (EU) 2016/679, the General Data Protection Regulation (GDPR, in Polish RODO): including data minimization, security of processing, recipients of data, automated decision-making (Art. 22) and data protection impact assessment (DPIA, Art. 35).
Who it concerns
Data controllers and processors, when documents, queries or AI results contain personal data.
AiTrust mechanism
- The range of documents available in work with AI depends on assigned permissions and source configuration.
- The model receives the context after sanitization of specified data categories.
- On-premise deployment in the customer’s infrastructure, when data should not go to an external model provider.
Evidence and the limits of what it confirms
The pre-send control panel shows which data categories were replaced with tags in the version passed to AI. This applies to one run and the specified categories; it does not prove that all personal data in other content is detected. Masking limits data disclosure but does not mean irreversible anonymization.
For the organization to decide
- The purpose and legal basis of processing and the scope of data approved for the given use.
- Data recipients, data processing agreements and the location of processing.
- Whether a data protection impact assessment (DPIA) is needed and how to treat decisions based on an AI result.
Next step
Identify the data categories in your process. In the demo, we will check which of them sanitization covers and where your organization needs to decide.
ISO/IEC 42001: AI management system in the organization
StandardSource of the requirement
ISO/IEC 42001 is an international standard that specifies the requirements for an artificial intelligence management system in an organization (AI management system, AIMS).
Who it concerns
Organizations that provide or use AI systems. Applying the standard and any certification are the organization’s decision.
AiTrust mechanism
- Deciding what requires human review and what content or action the approval of an AI result covers.
- Evidence trail for operations in the controlled path.
Evidence and the limits of what it confirms
Records from the tool can be one of the materials used in the management system. Neither the tool nor its log is a certification of the organization or the product; the management system is certified by a certification body.
For the organization to decide
- AI policy, roles and responsibilities.
- Assessment of the risk and impact of AI systems, objectives and reviews.
- Whether and when to seek certification of the management system.
Next step
Compare the scope of your management system with the control points in the process. In the demo, we will show what records are created in the controlled path.
Audit and documentation of work with AI
Organization’s rulesSource of the requirement
Documentation requirements come from the laws, standards, contracts and policies that apply in the given process.
Who it concerns
Process owners, internal audit, compliance and security teams, and the people who assess the run of work with AI.
AiTrust mechanism
- Evidence trail for operations in the controlled path: events, user, H.A.D.A. oversight zone and decisions.
- Log view with filters, export to CSV, Excel and PDF, and a function to verify the chain of records.
- The organization sets the evidence retention period.
Evidence and the limits of what it confirms
The run record describes a selected stage of work in the controlled path: who, when, what operation and what decision. It is not complete audit documentation, which also covers, among other things, the purpose of the use, the risk assessment, policies and decisions made outside the tool. Actions outside the configured paths are not recorded.
For the organization to decide
- The scope of recorded events and how long they are retained.
- Who has access to the records and who assesses their completeness: internal audit, an external auditor or a supervisory authority.
Next step
Decide who will assess the records and for what purpose. In the demo, we will show the log view and export for an example path.
Source of the requirement
Directive (EU) 2022/2555 (NIS2), including Art. 21 (cybersecurity risk-management measures) and Art. 23 (incident reporting). Obligations come from the national laws that implement the directive.
Who it concerns
Entities covered by the laws implementing NIS2, e.g. in the energy sector. Coverage depends on the sector, size and status of the entity.
AiTrust mechanism
- On-premise deployment in the customer’s infrastructure.
- Permissions to the documents and sources used in work with AI.
- Evidence trail for operations in the controlled path.
Evidence and the limits of what it confirms
The record of events in the AiTrust path can support the analysis of an event concerning that path. It does not cover the organization’s other systems and does not replace risk-management procedures or incident reporting.
For the organization to decide
- Whether the organization is covered by the laws and to what extent.
- Risk-management measures and procedures for handling and reporting incidents.
- Control of data sending on the organization’s devices and network.
Next step
Determine whether and how the laws cover your organization. In the demo, we will discuss a deployment option in your infrastructure.
Financial sector: KNF supervision and banking secrecy
Supervisory authorityNational lawExample in the sector FinanceSource of the requirement
The Polish Financial Supervision Authority (KNF) is a supervisory authority. Its recommendations, positions and communications, including those on processing information in cloud computing, are different documents, not a single law. Banking secrecy follows from banking law.
Who it concerns
Supervised entities, e.g. banks; the scope depends on the type of entity, the service and the supervisory document.
AiTrust mechanism
- On-premise deployment in the customer’s infrastructure.
- The model receives the context after sanitization of specified data categories.
- Operations of specified classes require a human decision before execution.
Evidence and the limits of what it confirms
The configuration and the run record can be presented in an internal assessment. Whether they meet supervisory expectations is assessed by the organization and, within its remit, by the supervisory authority.
For the organization to decide
- Which supervisory documents apply to the given process.
- Classification of any outsourcing and cloud services, and the risk assessment.
- Protection of data covered by banking secrecy in queries and results.
Next step
Identify the process and the supervisory documents that apply to it. In the demo, we will show the control points in an example path.
Professional secrecy in law firms
National lawProfessional ethicsSource of the requirement
The professional secrecy of advocates and attorneys-at-law follows from the laws on the legal professions and the ethics rules of the professional self-governing bodies.
Who it concerns
Law firms, legal departments and other teams working on documents covered by professional secrecy.
AiTrust mechanism
- The range of documents available in work with AI depends on assigned permissions and source configuration.
- The model receives the context after sanitization of specified data categories.
- On-premise deployment in the customer’s infrastructure with local models.
Evidence and the limits of what it confirms
The configuration shows where documents are processed and who has access to them. Sanitization of specified data categories does not remove all information covered by secrecy from a document.
For the organization to decide
- Which documents may go to which model and whether an external provider is acceptable.
- How to inform clients about the use of AI.
- Who approves the result before it is used.
Next step
Specify which case files and documents may be processed and where. In the demo, we will show work on a sample contract.
Source of the requirement
The GDPR classifies data concerning health as a special category of personal data (Art. 9). National laws on patients’ rights and professional secrecy also apply.
Who it concerns
Healthcare providers and other organizations that process patient data in documents and processes.
AiTrust mechanism
- On-premise deployment in the customer’s infrastructure with local models.
- The range of documents available in work with AI depends on assigned permissions and source configuration.
- The model receives the context after sanitization of specified data categories.
Evidence and the limits of what it confirms
The pre-send control panel shows which data categories were marked in the given operation. Sanitization covers the specified categories; other health information in the content may remain and requires assessment in the given process.
For the organization to decide
- The legal basis for processing special categories of data and the impact assessment (DPIA).
- Who uses the AI result; decisions on treatment are made by medical staff.
Next step
Choose an administrative or documentation process to check. In the demo, we will discuss the location of processing and the control points.
Source of the requirement
TISAX is an information security assessment scheme used in the automotive sector. It is not a law.
Who it concerns
Suppliers and partners of automotive companies whose customer requires a TISAX assessment.
AiTrust mechanism
- On-premise deployment in the customer’s infrastructure.
- The range of documents available in work with AI depends on assigned permissions and source configuration.
- Evidence trail for operations in the controlled path.
Evidence and the limits of what it confirms
A TISAX assessment result applies to the organization and the scope of the assessment, not to a single tool. The AiTrust configuration can be one of the elements described in that assessment.
For the organization to decide
- The assessment scope agreed with the partner.
- Classification of project data and of documentation received from customers.
Next step
Check the assessment scope your partner requires. In the demo, we will show permissions and the work record on project documents.
Law, supervision, standard or sector scheme
Requirements have different sources and different force. The source type is shown with each topic.
- EU law
- An EU regulation applies directly; an EU directive applies through the national laws that implement it. Obligations concern the entities within the scope of the law.
- National law
- National acts and regulations, e.g. on professional secrecy or patients’ rights.
- Supervisory authority
- Recommendations, positions and communications of a supervisory authority (e.g. KNF, the Polish Financial Supervision Authority) addressed to supervised entities. These are different documents, not a single law.
- Standard
- A voluntary standard (e.g. ISO/IEC 42001) that an organization implements and may have certified by a certification body.
- Sector scheme
- Requirements and assessments adopted in a given sector (e.g. TISAX in automotive), usually required by business partners rather than by law.
- Professional ethics
- Rules of the professional self-governing bodies, e.g. of advocates and attorneys-at-law.
- Organization’s rules
- The organization’s policies, procedures and contracts, including how work is documented.
AI Act application timeline
The provisions of the AI Act start to apply in stages. Which of them concern your organization depends on its role and the classification of the use.
- August 1, 2024The AI Act enters into force.
- February 2, 2025The general provisions (definitions and AI literacy) and the prohibitions apply.
- August 2, 2025The rules on general-purpose AI models apply; Member States designate the competent authorities and adopt the rules on penalties; governance bodies are set up at EU level.
- August 2, 2026Most provisions start to apply, including the transparency obligations (Art. 50), and their enforcement begins.
- December 2, 2026(date amended by the Digital Omnibus)New prohibitions on systems generating non-consensual sexual deepfake content and child sexual abuse material; a transitional deadline for some providers regarding Art. 50(2).
- August 2, 2027At least one AI regulatory sandbox operates in each Member State.
- December 2, 2027(date amended by the Digital Omnibus)The rules for high-risk AI systems listed in Annex III apply.
- August 2, 2028(date amended by the Digital Omnibus)The rules for high-risk AI systems embedded in products covered by Annex I apply.
Source European Commission, AI Act Service Desk, Timeline for the implementation of the EU AI Act, read on October 30, 2026. The page reflects the changes introduced by the Digital Omnibus on AI. Before making a decision, check the current status on the source page.
Methodologies and control examples
Methodologies give structure to the organization’s decisions, and control examples show how AiTrust watches over them while AI digital workers do their work.
Methodologies
- B.O.R.D.E.R.AI operating boundariesOpen the methodology
- H.A.D.A.Human oversight of AIOpen the methodology
- I.A.G.R.A.Readiness assessment methodologyOpen the methodology
- S.H.I.E.L.D.Protection and resilienceOpen the methodology
Control examples
- Boundaries in the workflowSee the example
- Operation risk assessmentSee the example
- Controlled executionSee the example
- Accountability for decisionsSee the example
- Documenting the workflowSee the example
- Resilience and failure handlingSee the example
Let’s check the requirements using your process as an example.
We will show which AiTrust mechanisms apply to your use and what work record is created in the controlled path. Your organization carries out the legal and organizational assessment.