Block Mondo designs and operates AI-enabled capabilities so organisations can adopt AI progressively while maintaining clear accountability, appropriate human oversight and practical controls over data, models, agents and automated actions. Our approach is framework-based, AI-driven and human-led.
1. Purpose
This statement explains the principles Block Mondo applies when designing, configuring, deploying and operating AI-enabled features across the Block Mondo platform. It is intended to provide customers, users and other stakeholders with a clear public summary of our responsible AI position.
The statement is derived from the MondoAI Product Strategy and the wider Block Mondo governance model. It does not represent an independent certification or accreditation.
2. Our responsible AI principles
| Principle | Block Mondo commitment |
|---|---|
| Use-case accountability | AI use cases have a defined business purpose, named ownership and an approval state. |
| Risk proportionality | The level of assessment, testing, approval and human oversight should increase with the potential impact and risk of the use case. |
| Human oversight | Material actions can be subject to explicit human review, approval steps and configured authority limits. |
| Transparency | Where appropriate, AI interactions include indicators, source references, version information and audit metadata so users can understand how an AI-enabled outcome was produced. |
| Privacy and data governance | AI features use approved knowledge and data sources, respect user and tenant permissions, and apply data minimisation, retention and lifecycle controls. |
| Testing and evaluation | AI-enabled capabilities are evaluated against defined scenarios, acceptance criteria and regression checks before material changes are released. |
| Contestability and escalation | AI-supported outcomes should be reviewable, with escalation and human intervention available where the use case requires it. |
| Accountability and auditability | Configuration changes, relevant user activity, agent actions and tool execution can be recorded to support investigation and accountability. |
3. Human-led AI
Block Mondo does not treat AI as a replacement for organisational accountability. AI may assist, recommend, prepare or execute bounded actions depending on the approved use case, but responsibility for material business decisions remains with the organisation and its authorised people.
- Human approval checkpoints can be inserted before material actions.
- Agent authority can be constrained to read, draft, recommend, prepare or execute specific approved actions.
- Users are expected to apply appropriate judgement before relying on AI-generated outputs.
- Business-critical deterministic rules should remain in normal application logic rather than being delegated to a probabilistic model.
4. Governed AI models, agents and workflows
MondoAI provides a common governance layer for the AI capabilities used across Block Mondo applications. Rather than allowing individual applications to integrate AI models independently, model selection, knowledge, agents, workflows, permissions, evaluation and monitoring can be governed centrally.
- Approved model and provider catalogues support controlled model selection.
- Provider-specific data handling and contractual controls can be recorded centrally.
- Agents have defined roles, objectives, instructions, knowledge sources, tools, authority and escalation behaviour.
- Tools and integrations are allow-listed and use explicit service contracts and permissions.
- Prompts, agent instructions and workflow configuration are versioned.
- Long-running or multi-step AI workflows retain state and audit information where appropriate.
5. Data, privacy and knowledge
AI is only useful when it operates with trustworthy organisational context. Block Mondo's approach is to use governed knowledge and approved data sources while preserving provenance, permissions and lifecycle controls.
- Knowledge retrieval honours the initiating user's application and tenant permissions.
- Sources can retain metadata such as owner, classification, version and provenance.
- Where appropriate, knowledge-grounded answers can provide citations or source links.
- Knowledge can be re-indexed, expired or removed as source information changes.
- Data minimisation and retention controls are applied according to the use case and customer requirements.
- Customer-specific privacy, residency and provider requirements are considered when selecting models and services.
Users also have responsibilities when using AI-enabled features. Block Mondo will make every effort to detect private or sensitive information before it is shared with an AI model where relevant safeguards are enabled, but automated detection cannot be guaranteed to identify every instance. Users should not submit private, sensitive, secret or regulated information unless the feature, their organisation and applicable instructions expressly permit it.
6. AI agents and personal information
Where an AI agent retains notes, context, check-ins or other personal information, the relevant consent or acknowledgement is presented within the platform. Private AI-agent experiences are designed so that personal interactions are not exposed as general organisational reporting content. Where aggregated outcome or theme reporting is enabled, appropriate minimum-group thresholds and disclosure controls can be applied.
7. Testing, monitoring and change control
Responsible AI is an operational discipline rather than a one-time design decision. Block Mondo uses evaluation, controlled release and monitoring practices to manage AI-enabled capabilities through their lifecycle.
- Evaluation scenarios and acceptance criteria are defined for material use cases.
- Regression evaluation can be performed before material model, prompt or workflow changes.
- Usage, quality, latency, cost, exceptions and human overrides can be monitored where practical.
- Models, knowledge, agents and workflows are retired or replaced through controlled lifecycle processes.
- Incidents, user feedback and identified quality issues can feed into remediation and improvement.
8. Model choice and third-party providers
Block Mondo is model-agnostic at the platform level. Approved providers and models may be selected according to capability, privacy, residency, cost, latency and risk. A customer may also have provider-specific requirements under its agreement or configuration.
Third-party AI models remain independent services. Block Mondo governs how approved models are selected and used within the platform, but does not represent that it controls or guarantees every behaviour of an independent model provider.
9. Governance and assurance
Responsible AI controls are designed to support consistent governance across the platform and to provide evidence for customer, regulatory and assurance requirements. Read about our governance alignment and assurance approach in the Trust Centre.
10. Continuous improvement
AI technology, regulation and accepted practice continue to evolve. Block Mondo will review and update its responsible AI controls as the platform changes, new risks emerge, customer requirements develop and relevant standards or regulatory expectations change.