---
title: "AI Act Assessment for Convier Customers"
visibility: "public"
---

> This section collects the materials most often requested during assessment of Convier under the AI Act

## 1. Purpose of this Document

This document describes how Convier AI Agents are designed to support governance, control, transparency, traceability, and responsible use of artificial intelligence within financial crime and compliance workflows.

The document is intended to support Customers in their assessment and governance of Convier AI Agents under Regulation (EU) 2024/1689, the Artificial Intelligence Act.

This document may also support assessments relating to GDPR, internal AI policies, information security requirements, outsourcing requirements, operational resilience, and other governance or risk management requirements applicable to the Customer’s use of AI.

Convier will support the Customer by providing product information, technical documentation, configuration information, operational evidence, and reasonable assistance relevant to the Customer’s AI governance and assessment processes.

<Callout type="warning" title="Legal and regulatory determinations">

This document is not legal advice. The final legal and regulatory determinations for the Customer’s own use of AI remain with the Customer.

The classification and obligations applicable to an AI system depend on the specific intended use, configuration, workflow, degree of automation, user role, data processed, and operational context.

</Callout>

## 2. Executive Summary

Convier AI Agents are designed to operate within configured Convier workflows. They are not general-purpose AI assistants operating independently from the Customer’s Convier configuration.

The AI Agents service is designed to support and automate tasks within licensed workflows, including customer due diligence, ongoing monitoring, transaction monitoring investigations, reporting, and related financial crime prevention processes.

Convier’s approach is based on the following principles:

- AI agents operate within existing workflows configured in Convier;
- AI agents operate within the same Convier environment as authorised users and use the configured data integrations, workflows, controls, analysis functionality, and reporting mechanisms made available within the Customer’s Convier deployment;
- AI agents are governed through the Control Tower;
- the Customer’s own AI environment is used, such as Azure AI Foundry, Azure OpenAI, Google’s enterprise AI platform, or another approved enterprise AI environment;
- the Customer selects and approves the AI models and AI services used;
- agent tasks, data access, prompts, output formats, automation levels, and review requirements are configured within the agreed scope;
- human oversight and human-in-the-loop or human-on-the-loop review can be configured at workflow, task, and agent level;
- AI-generated and AI-assisted output can be logged and traced;
- confidence indicators and manual review flags can be used to prevent automatic completion where the agent identifies uncertainty;
- data minimisation, filtering, masking, redaction, or pseudonymisation may be configured on a best-efforts basis;
- Customer Data may be used for testing, evaluation, adaptation, fine-tuning, training, or improvement where expressly agreed; and
- Convier will provide documentation, configuration visibility, audit trails, and evidence to support the Customer’s AI governance and regulatory assessments.

<Callout type="info" title="Shared responsibility">

Convier provides the operating environment, orchestration layer, governance controls, documentation, and evidence needed to support controlled use of AI agents. The Customer provides the organisational context, internal procedures, model approval, legal basis, and final legal and regulatory determinations for its own use of AI.

</Callout>

## 3. Description of Convier AI Agents

An AI agent is a configurable software component within the Convier platform that uses artificial intelligence to perform predefined tasks within an approved workflow configured in Convier.

AI agents operate within the same Convier environment as authorised users and use the configured data integrations, workflows, controls, analysis functionality, and reporting mechanisms made available within the Customer’s Convier deployment.

Depending on the agreed scope and configuration, an AI agent may:

- retrieve and structure relevant information from Customer-designated data sources;
- summarise customer, transaction, case, or workflow-related information;
- compare available information against workflow requirements, procedures, policies, routines, or expected customer behaviour;
- identify missing information, inconsistencies, or matters that require review;
- generate draft narratives, reports, comments, assessments, or investigation summaries;
- produce structured output in accordance with the configured task instruction and output format;
- recommend, initiate, or complete workflow actions where configured and approved;
- flag uncertainty, insufficient information, or conflicting information for manual review; and
- support execution and documentation of configured workflow processes.

AI agents operate within the workflows, permissions, data sources, instructions, controls, and review requirements configured for the Customer in Convier.

## 4. Architecture and Technical Setup

Convier AI Agents are designed to operate within the Customer’s technical environment and in connection with the Customer’s existing Convier deployment.

Unless otherwise agreed, the AI agents use the Customer’s own AI environment, such as Azure AI Foundry, Azure OpenAI, Google’s enterprise AI platform, or an equivalent AI service environment approved by the Customer.

The Customer selects, provides, configures, and maintains access to the relevant large language models and AI services used by the AI agents. These models will typically be GPT-family models or other models made available in the Customer’s approved AI environment.

Convier operates as an orchestration, workflow, and governance layer for the Customer’s selected AI environment. Convier does not normally operate a proprietary language model.

Convier orchestrates the AI agent workflow through the Convier platform. This may include:

- preparing task instructions;
- retrieving relevant data from approved sources;
- applying configured data minimisation, filtering, masking, redaction, or pseudonymisation measures;
- sending prompts and relevant context to the Customer’s selected AI model or AI service;
- instructing the model to produce output in a configured structure;
- receiving model output;
- presenting the output to users or workflow components;
- applying configured workflow rules, confidence indicators, and manual review flags; and
- logging relevant activity.

Unless otherwise agreed, Convier does not provide the underlying foundation AI model, operate the Customer’s AI environment, enter into the Customer’s agreement with the relevant AI model or AI service provider, decide which models the Customer may use, or control the terms on which the Customer’s selected AI model or AI service processes data.

## 5. Intended Use and AI Act Classification

<Callout type="info" title="Intended use and classification">

The AI Act classifies AI systems based on their intended use. The relevant question is therefore not only what Convier AI Agents are technically capable of doing, but how a specific Customer configures and uses them in practice.

</Callout>

Convier AI Agents are intended to support and automate tasks within configured workflows. Convier will support the Customer in documenting the intended use of each AI agent, including:

- the workflow in which the agent is used;
- the task or action the agent performs;
- the data sources and datasets available to the agent;
- the output the agent may produce;
- the degree of automation permitted;
- whether the agent may initiate, recommend, or complete workflow actions;
- whether human review is mandatory or optional;
- what happens if the agent identifies uncertainty or insufficient information;
- whether the output affects customer onboarding, customer treatment, customer risk classification, reporting, investigation outcomes, or access to financial services; and
- which internal procedures, controls, and governance requirements apply.

Based on Convier’s current standard functionality for financial crime workflows, Convier does not consider the AI Agents service to fall within the prohibited AI practices set out in Article 5 of the AI Act.

For financial institutions, the most important assessment is whether the AI system is used in a way that makes or materially influences decisions affecting natural persons, including decisions relating to access to financial services, creditworthiness, customer treatment, customer risk assessment, or similar outcomes.

Use of AI in AML, fraud, or compliance processes should not automatically be treated as high-risk merely because it concerns financial crime prevention. The assessment depends on the specific intended use and whether the system falls within an AI Act high-risk category or materially influences a relevant decision.

The AI Act distinguishes between administrative analysis of information in an AML context and AI systems used for law enforcement, criminal investigation, prosecution, or judicial decision support. Banks and other obliged entities perform their own obligations under AML legislation. They are not law enforcement authorities. Use of AI in a bank’s AML work should therefore not automatically be treated as use of AI for law enforcement purposes.

If an AI agent is used only to support narrow procedural tasks, information retrieval, summarisation, drafting, or preparation of material for human review, this may support an assessment that the AI system does not materially influence the outcome of decision-making. However, that assessment must be made based on the concrete configuration and operational use.

If an AI agent is configured to autonomously initiate, complete, recommend, or materially influence decisions or workflow actions with significant legal, regulatory, financial, or operational consequences for natural persons, the Customer should perform a more detailed high-risk assessment. Convier will support such assessment by providing configuration information, workflow descriptions, task instructions, automation settings, logging information, and other relevant evidence available in the platform.

A future or customer-specific configuration may require separate assessment if AI agents are used to create alerts, propose or assign customer risk classifications, escalate cases automatically, recommend measures towards customers, or otherwise materially affect decisions concerning customers.

This does not mean that such a use case is automatically high-risk. It means that the classification assessment should be based on the system’s intended purpose, actual operational role, degree of autonomy, and the significance of AI output for decisions affecting customers.

## 6. Governance Model and Responsibilities

Convier and the Customer work together to support responsible and controlled use of AI agents within the Customer’s Convier deployment.

Convier provides the operating environment, orchestration layer, configuration functionality, Control Tower, logging, traceability, human oversight functionality, and documentation support for AI-supported financial crime workflows.

The Customer provides the organisational and regulatory context in which the AI agents are used, including the relevant workflows, internal procedures, risk appetite, approval requirements, model selection, and legal or regulatory requirements applicable to its organisation.

### 6.1 Convier Support

Convier will support the Customer by providing relevant information, documentation, configuration support, implementation support, and reasonable assistance relating to:

- AI agent configuration;
- workflow controls;
- Control Tower configuration;
- human oversight functionality;
- logging and traceability;
- data minimisation, filtering, masking, redaction, or pseudonymisation configuration;
- structured output formats;
- confidence indicators and manual review flags;
- technical integration with the Customer’s AI environment;
- use of the Customer’s selected AI model or AI service within Convier workflows;
- testing, validation, and review of configured AI agent functionality; and
- operation of AI agents in Convier.

Convier will also provide information and evidence reasonably available in the Convier platform to support the Customer’s AI governance, internal control, audit, risk assessment, data protection assessment, and AI Act assessment processes.

Such support may include information about system architecture, data flow, configured workflows, AI agent instructions, data sources, output formats, Control Tower settings, logging, review steps, and AI-generated or AI-assisted outputs.

### 6.2 Customer Input and Decisions

The Customer provides the organisational and regulatory input required to configure and govern the AI agents appropriately.

This includes input on the intended use of the AI agents, applicable workflows, internal procedures, permitted data sources, model or AI service provider, permitted degree of automation, human oversight requirements, access rights, retention requirements, and relevant internal governance requirements.

The Customer makes the final determinations regarding its own legal, regulatory, operational, AI governance, data protection, and compliance obligations, including the classification of its intended use under the AI Act.

Convier’s support is intended to assist the Customer in making and evidencing those determinations, but does not constitute legal, regulatory, financial, or compliance advice.

### 6.3 AI Model or AI Service Provider

Where the Customer uses a third-party AI model or AI service, the Customer’s relationship with the provider is governed by the Customer’s agreement with that provider.

Convier will configure and govern the use of the selected AI model or AI service within the agreed Convier workflows, including through task instructions, data access rules, output formats, Control Tower configuration, logging, and human oversight functionality.

The Customer and Convier will cooperate to ensure that the selected model or AI service is used in accordance with the agreed technical setup, the Customer’s instructions, and the applicable workflow configuration.

## 7. Control-Level Support for AI Act Governance

Convier is designed to support operational governance of AI agents. This means that the Customer’s procedures, controls, data requirements, review requirements, and escalation rules can be reflected directly in how AI agents are configured and used in Convier, rather than only being described in separate policy documents.

The following sections describe the main control areas supported by Convier AI Agents, depending on the Customer’s licensed workflows, technical environment, AI model configuration, and agreed implementation scope.

### 7.1 Intended Use and Scope Control

AI agents operate only within configured Convier workflows.

The Customer and Convier define which workflows, tasks, data sources, datasets, output formats, automation levels, and review requirements apply to each AI agent.

This supports documentation of intended use and prevents AI agents from operating as general-purpose tools outside the agreed workflow scope.

Convier will provide configuration information and workflow documentation to evidence the scope within which each AI agent is configured to operate.

### 7.2 Control Tower Governance

Convier distinguishes between AI execution and AI governance. AI agents perform operational tasks, while the Control Tower defines the procedures, controls, data requirements, output requirements, escalation rules, and review requirements that govern how those tasks are performed.

The Control Tower is the governance and configuration layer for AI agents in Convier.

Depending on configuration, it can be used to define, review, and manage:

- the tasks each AI agent is configured to perform;
- the tools, functions, or workflow actions each AI agent may use or perform;
- the data sources and datasets each AI agent may access;
- which data sources, datasets, or information elements are required before a task may be completed;
- the prompt, task instruction, or agent instruction used by each AI agent;
- the policies, procedures, routines, or control requirements the AI agent should take into account;
- the selected AI model or AI service used by the AI agent;
- the output format and documentation structure;
- rules for handling insufficient, incomplete, or conflicting information;
- required human review, approval, rejection, escalation, or override steps;
- logging, audit trail, and traceability requirements;
- confidence indicators showing the AI agent’s assessment of whether a task has been performed in line with the applicable instruction, including manual review flags where the agent identifies uncertainty, insufficient information, or other circumstances preventing automatic completion; and
- user access rights and permissions relating to AI agent functionality.

The Control Tower supports the Customer’s ability to align AI agent configuration with internal procedures, workflow requirements, human oversight requirements, and control frameworks.

Control Tower is intended to ensure that AI agents do not merely produce an answer whenever they are asked to perform a task. The agent should perform the task only when the required conditions are satisfied. If required information is missing, conflicting, or insufficient, the agent can be configured to stop, escalate, or return the task for manual review instead of producing an unsupported conclusion.

Convier will provide Control Tower configuration information and related logs to evidence how AI agents are instructed, constrained, reviewed, and governed.

### 7.3 Human Oversight, Human-in-the-Loop, and Human-on-the-Loop Controls

Convier AI Agents include functionality to support human oversight and human-in-the-loop or human-on-the-loop review.

Depending on configuration, authorised users can review AI-generated output, edit or supplement output, approve or reject output, override AI-initiated actions, request re-runs, view relevant information used or referenced by the AI agent where available, document comments and reasoning, and maintain an audit trail of AI agent activity and user actions.

Human oversight may include mandatory review steps, role-based approvals, escalation points, restrictions on autonomous execution, ongoing sample checks, quality control, or rules preventing automatic completion where confidence is insufficient.

Convier supports several oversight models:

| Oversight model | Description |
| --- | --- |
| Human-in-the-loop | Human approval is required before a specific action is completed. |
| Human-on-the-loop | AI agents perform tasks autonomously, while human users monitor results, review activity, perform quality control, and intervene where necessary. |
| Autonomous execution within defined boundaries | AI agents perform predefined tasks autonomously within approved authority limits, data requirements, escalation rules, logging requirements, and review mechanisms. |

The appropriate oversight model depends on the use case, degree of autonomy, significance of the task, risk assessment, and the Customer’s internal procedures.

Convier will provide configuration information, logs, review history, approvals, overrides, rejections, and escalation records to support evidence of human oversight and review.

### 7.4 Confidence Indicators and Manual Review Flags

Convier AI Agents can include confidence indicators showing the agent’s assessment of whether a task has been performed in line with the applicable instruction, output format, and workflow requirements.

Where the agent identifies uncertainty, insufficient information, conflicting information, or other circumstances preventing automatic completion, the agent may render a manual review flag.

<Callout type="warning" title="Manual review flag limitations">

Manual review flags are intended to support human oversight and quality control. They do not guarantee that all errors, uncertainties, or issues will be identified.

</Callout>

Convier will make confidence indicators and manual review flags visible in the workflow and, where configured, available in logs or workflow history.

The Customer and Convier can define how confidence indicators and manual review flags are used in each workflow, including whether they prevent automatic completion, require approval, or trigger escalation.

### 7.5 Logging, Traceability, and Audit Trail

Convier provides functionality to support logging and traceability of AI agent activity.

Depending on configuration, logs may include:

- the AI agent used;
- the workflow, case, customer, alert, or task in which the AI agent was used;
- the user initiating, reviewing, approving, rejecting, editing, or overriding the output;
- relevant timestamps;
- task instructions, prompts, or prompt references;
- data sources accessed or referenced;
- controls performed;
- information analysed;
- observations generated;
- actions performed;
- conclusions produced;
- the selected AI model or AI service used;
- generated output;
- structured output fields;
- confidence indicators or manual review flags;
- user edits, approvals, rejections, comments, or overrides; and
- system events related to AI execution.

Logs are written to the logging solution configured by Convier in collaboration with the Customer or to another Customer-designated logging solution as agreed.

Convier will provide AI-related logging functionality and available audit trails to support review, investigation, governance, and audit processes.

The Customer uses such logs in accordance with its internal policies, regulatory obligations, legal retention requirements, and governance framework.

### 7.6 Transparency and AI-Generated Output

Convier supports transparency by making AI-generated or AI-assisted output visible within the workflow.

Depending on configuration, the platform can distinguish between:

- AI-generated content;
- AI-assisted content;
- user-edited content;
- user-approved content; and
- final workflow output.

This supports the Customer’s ability to understand, review, document, and evidence how AI-generated output has been used.

AI activity should be visible, traceable, and controllable. Users and control functions should be able to understand where AI has been used, which tasks were performed by AI, which outputs were generated or supported by AI, and how the output was produced.

Convier will provide workflow views, output history, logs, and review information to support this transparency.

### 7.7 Data Governance and Prompt Context Control

Convier AI Agents operate on Customer-designated data sources and within configured access rights and workflow scope.

The AI agent configuration can limit which data sources are available to an agent, what context is retrieved, and what information is included in prompts sent to the Customer’s selected AI model or AI service.

The Control Tower supports purpose limitation and data minimisation by enabling the Customer and Convier to define which data sources may be used, which data sources must be used, and which information is required for a particular task.

Convier can support configured data minimisation, filtering, masking, redaction, or pseudonymisation of selected Customer Data before such data is included in prompts or contextual input sent to the Customer’s selected AI model or AI service.

<Callout type="warning" title="Data minimisation limitations">

Such functionality is provided on a best-efforts basis and is intended to reduce the amount and sensitivity of information made available to the selected AI model or AI service for the relevant task.

Redaction, masking, filtering, or pseudonymisation may not identify or remove all personal data, sensitive information, confidential information, or information capable of identifying an individual or entity in all circumstances. Information that has been masked, redacted, or pseudonymised may still be capable of being inferred, reconstructed, or linked to an individual or entity when combined with other data, contextual information, transaction patterns, relationship data, or external sources.

</Callout>

Convier will provide configuration information, data source information, and available logs showing relevant data sources accessed by AI agents. The Customer uses this information to support its assessment of which data may lawfully be made available to an AI agent or AI model, including any requirements relating to personal data, confidential information, special categories of personal data, information relating to criminal convictions and offences, banking secrecy, professional secrecy, or other protected information.

### 7.8 Access Control and Workspace Separation

Convier AI Agents operate subject to configured user permissions, identity and access management, workflow permissions, and environment setup.

Depending on configuration, access to AI agents, workflows, cases, data sources, output, logs, and Control Tower functionality can be restricted by role, user group, workflow, environment, or other access control mechanisms.

Convier will support role-based configuration, workflow permissions, and access control setup within the Convier platform.

The Customer provides the relevant organisational access model, user groups, identity and access management setup, and internal approval requirements.

### 7.9 Monitoring, Incident Handling, and Change Management

Convier supports AI agent monitoring, incident handling, and change management within Convier’s scope of responsibility and in accordance with the Agreement.

Depending on configuration, relevant AI-related events may be logged and made available to support investigation, review, incident assessment, and governance processes.

The Customer can monitor agent activity, limit the agent’s authority, override outputs, stop execution, review historical results, and update the procedures governing how AI tasks are performed, depending on agreed scope and configuration.

Convier will provide support, logs, support records, incident information, and remediation information within Convier’s scope to support assessment and follow-up of AI-related events.

The Customer makes the final determination of whether an AI-related event constitutes an internal incident, regulatory incident, security incident, data protection incident, serious incident, or other reportable event under applicable law.

Where an AI-related incident involves the Convier platform, Convier and the Customer cooperate in accordance with the Agreement, including applicable support, security incident, notification, and governance provisions.

## 8. Mapping to AI Act Requirements

The following table summarises how Convier AI Agents can support the Customer’s AI Act governance and control framework. The applicability of each requirement depends on the classification and intended use of the AI system.

| AI Act area | Relevance | Convier control | Evidence provided |
| --- | --- | --- | --- |
| Classification and intended use | AI Act classification depends on intended use and whether the system falls within a prohibited or high-risk category. | Convier configures AI agents for defined workflows, tasks, data sources, outputs, automation levels, and review requirements. | Workflow configuration, AI agent configuration, task instructions, Control Tower setup, implementation documentation, and intended-use descriptions. |
| Prohibited AI practices | Certain AI practices are prohibited. | Based on current standard financial crime functionality, Convier does not consider the AI Agents service to fall within Article 5 prohibited practices. Customer-specific uses can be assessed separately. | Description of current functionality, configured workflows, task scope, data sources, automation settings, and customer-specific configuration documentation. |
| Risk management | High-risk AI systems require lifecycle risk management. | Convier provides Control Tower configuration, workflow controls, human oversight settings, confidence indicators, manual review flags, logging, monitoring, and change management support. | Control Tower setup, review rules, escalation rules, confidence indicators, manual review flags, change history, logs, and test/validation material. |
| Data governance | Data used by AI systems should be appropriate for the intended purpose. | Convier supports approved data sources, required data source rules, workflow-specific context retrieval, data minimisation, masking, redaction, pseudonymisation, and customer-defined data access rules. | Data source configuration, access rules, prompt-context configuration, redaction or masking configuration, data-flow description, and logs showing data sources accessed. |
| Technical documentation | Providers and deployers may need documentation of system design, intended use, controls, and operation. | Convier will provide product information, architecture documentation, implementation documentation, workflow configuration information, and reasonable support for Customer assessments. | Product documentation, architecture descriptions, implementation documentation, workflow configuration, task instructions, and support material. |
| Record-keeping | High-risk systems must support logging and traceability. | Convier provides logging of AI agent activity, prompts or prompt references, data sources, model used, output, confidence indicators, manual review flags, controls performed, and user actions, depending on configuration. | Audit logs, workflow history, AI activity logs, prompt references, model references, generated output, user approvals, overrides, rejections, and escalation records. |
| Transparency | Deployers must be able to interpret outputs and use systems appropriately. | Convier provides structured output, visible AI-generated or AI-assisted content, workflow context, source references where available, and review interfaces. | AI-generated output, structured output fields, source references where available, user review history, edits, approvals, and final workflow output. |
| Human oversight | High-risk AI systems must enable human oversight proportionate to risk, autonomy, and context. | Convier supports human-in-the-loop, human-on-the-loop, approval, rejection, override, escalation, restrictions on autonomous completion, and manual review flags. | Review configuration, approval rules, role permissions, manual review flags, user actions, override history, escalation records, and quality-control records. |
| Accuracy, robustness, and cybersecurity | AI systems must be designed and operated with appropriate robustness and cybersecurity controls. | Convier supports deployment in the Customer environment, role-based access, workflow permissions, logging, audit trail, configuration controls, confidence indicators, manual review flags, and use of the Customer’s security architecture. | Access control configuration, environment setup, logging configuration, audit trail, workflow permissions, testing records, and incident/support records. |
| Deployer obligations | Deployers must use systems according to instructions, assign oversight to competent persons, monitor operation, and keep logs where under their control. | Convier provides configuration, instructions, logs, Control Tower, support, and review functionality. | User permissions, Control Tower configuration, AI agent instructions, usage logs, review logs, monitoring records, and training records maintained by the Customer. |
| Incident handling | Serious incidents may require assessment and reporting. | AI-related events can be logged and reviewed. Convier supports incidents within its scope. | Incident logs, support tickets, security incident notifications where applicable, workflow history, affected outputs, user actions, and remediation records. |
| General-purpose AI model obligations | Underlying general-purpose AI models may be subject to separate obligations. | The Customer selects and approves the model or AI service provider. Convier configures and governs use of the selected model within Convier workflows. | Model configuration, provider selection, customer approval records, AI environment setup, model references in logs, and provider documentation held by the Customer. |

## 9. High-Risk Assessment Considerations for Financial Crime Use Cases

Convier AI Agents may be used in workflows related to AML, financial crime, fraud, customer due diligence, ongoing monitoring, and transaction monitoring investigations.

Convier will support the Customer’s high-risk assessment by providing information about the configured AI agents, workflow role, data sources, task instructions, output formats, automation levels, human oversight settings, and audit trails.

Relevant assessment questions include:

- Is the AI agent used to evaluate the creditworthiness of natural persons or establish a credit score?
- Is the AI agent used to determine or materially influence access to financial services?
- Is the AI agent used to make or materially influence customer onboarding, rejection, termination, freezing, blocking, or other customer treatment decisions?
- Is the AI agent used to independently classify a customer as high risk, suspicious, or unsuitable?
- Is the AI agent used to autonomously submit reports to authorities?
- Is the AI agent used to create transaction monitoring alerts or materially influence alert generation?
- Is the AI agent used only to support narrow procedural tasks, information retrieval, summarisation, drafting, or preparation for human review?
- Does the AI agent’s output materially influence the outcome of decision-making?
- What degree of human review, approval, override, or escalation applies?
- What authority limits and workflow controls apply?
- What are the consequences if the AI agent output is incorrect, incomplete, or unsuitable?

The same AI agent technology may support different classifications depending on how it is configured and used.

## 10. Article 50 Transparency Considerations

Where AI-generated text or other output is produced, transparency obligations may be relevant depending on the use case and audience.

Convier supports transparency by enabling AI-generated or AI-assisted output to be presented within the workflow and, depending on configuration, distinguished from user-generated or user-approved content.

Convier will provide functionality and evidence to support the Customer’s assessment of how AI-generated content is produced, reviewed, edited, approved, or used in final workflow output.

Where required, the Customer determines whether and how AI-generated content should be disclosed, labelled, marked, reviewed, or communicated to users, customers, authorities, or other third parties.

## 11. AI Literacy and User Training

The Customer should ensure that users involved in the operation, review, approval, or oversight of AI agents have appropriate knowledge and competence.

Convier will support AI literacy and user training by providing product guidance, documentation, walkthroughs, and reasonable support relating to the use of AI agents in Convier.

Relevant training topics may include:

- what the AI agent is configured to do;
- what data the AI agent may access;
- which tools, functions, or workflow actions the agent may use or perform;
- how the agent’s output should be interpreted;
- limitations of AI-generated output;
- risks of over-reliance;
- how to review, approve, reject, or override output;
- how confidence indicators and manual review flags should be interpreted;
- how to document review and reasoning;
- when to escalate issues; and
- applicable internal policies and procedures.

The Customer incorporates such information into its internal training and AI literacy programme.

## 12. GDPR and Data Protection Considerations

AI-supported work with financial crime prevention may involve processing personal data, including customer information, transaction data, ownership information, adverse media, case notes, and other information relevant to AML, fraud, and financial crime prevention.

Convier supports the Customer’s assessment of data protection implications in several ways.

First, Convier is normally deployed in the Customer’s existing cloud environment, and AI functionality is normally activated through a model platform approved by the Customer. Convier will provide information about the relevant architecture, data flow, AI integration, and use of the selected AI environment.

Second, the Control Tower supports purpose limitation and data minimisation by enabling the Customer and Convier to define which data sources may be used, which data sources must be used, and which information is required for a particular task. This helps ensure that agents are not given unrestricted access to Customer information.

Third, the Control Tower supports accuracy and accountability by enabling the Customer and Convier to define data requirements, rejection criteria, and escalation rules. If necessary information is unavailable, the agent can be instructed to stop or escalate rather than produce an unsupported conclusion.

Fourth, Convier supports transparency and auditability by showing where AI has been used, which information has been used, which output has been generated, and whether a human has reviewed, changed, approved, or rejected the output.

Fifth, Convier supports human oversight and the possibility of intervention. This is relevant both for AI governance and for assessment of whether a use case may involve automated decision-making with legal effects or similarly significant effects for individuals.

Convier will provide available documentation and evidence relevant to DPIA assessments, including information about system architecture, data flow, data sources, access rights, logging, output handling, retention configuration, model integration, and human oversight functionality.

The Customer uses this information to assess and document:

- lawful basis;
- permitted purposes for AI use;
- categories of personal data that may be processed;
- whether special categories of personal data or information relating to criminal convictions and offences may be involved;
- the roles of Convier, Microsoft, Google, and any other relevant parties;
- controller, processor, and sub-processor relationships;
- access management;
- retention periods for prompts, outputs, logs, and case documentation;
- security measures;
- internal approval requirements; and
- procedures for review and challenge of AI-supported outputs.

Convier supports these assessments by making AI configuration, data use, execution, output, and oversight visible, reviewable, and auditable.

## 13. Customer Data, Model Use, and Training

Convier AI Agents may process Customer Data to perform tasks within the relevant workflow and agreed scope.

Customer Data may be used to configure, test, evaluate, adapt, fine-tune, train, or improve AI agents, AI models, prompts, retrieval methods, or related AI functionality where expressly agreed between the Parties and carried out in accordance with the Agreement, applicable data processing terms, Customer instructions, and applicable law.

Where the Customer uses a third-party AI model or AI service, the processing of Customer Data by that provider is governed by the Customer’s agreement with the provider and the provider’s applicable terms.

Convier will provide information about configured data use, data sources, model integration, prompt context, training or evaluation activities within the agreed scope, and related controls to support the Customer’s governance and documentation.

## 14. Documentation and Evidence Provided by Convier

Convier will make relevant documentation, configuration information, logs, and operational evidence available to support the Customer’s AI governance and assessment processes.

Technical architecture documentation is made available through Convier’s documentation pages, including [docs.convier.com](/docs).

Information about configured data sources, datasets, workflows, and AI agent configuration is available in the Customer’s Convier instance, including through the relevant workflow configuration and Control Tower functionality.

AI-related logs are written to the Customer’s designated logging solution, in the same manner as other Convier logs, unless otherwise agreed. The Customer can use such logs together with platform activity, workflow history, and Control Tower configuration to review and evidence AI agent activity.

Depending on the agreed scope and configuration, relevant evidence may include the AI agent configuration, task instructions, data sources, required data elements, model or AI service configuration, workflow actions, human review steps, confidence indicators, manual review flags, generated output, user actions, and logs.

Convier will support the Customer’s AI Act-related assessments by making AI configuration, data use, execution, output, review, and oversight visible, reviewable, and auditable through the Convier platform, the Customer’s configured logging solution, and Convier’s documentation.

The Customer uses this information as part of its internal governance, legal, regulatory, data protection, risk management, and audit processes.

## 15. Legal and Regulatory Assessment

The AI Act classification and compliance obligations applicable to Convier AI Agents depend on the Customer’s intended use, configuration, operational context, degree of automation, data use, and role under applicable law.

Convier will support the Customer’s assessment by providing relevant product information, technical documentation, configuration information, logs, audit trails, workflow evidence, and reasonable assistance within the agreed scope.

The Customer makes the final legal and regulatory determinations for its own use of AI agents, including its role under the AI Act, whether the intended use is high-risk, whether a fundamental rights impact assessment or DPIA is required, whether an incident is reportable, and whether a specific use of AI is lawful.

Convier’s role is to provide the technology, controls, documentation, and operational evidence needed to support that assessment and to help the Customer govern AI agent use in a controlled, traceable, and reviewable manner.

---

Source reference: [Original Notion page](https://app.notion.com/p/38a757c9c98c8059acbfe975575c745a)
