Repository navigation
04. Legal & Ethical Issues
Because Orbit is designed for the financial advisory sector and may process sensitive client information, privacy and data protection are important considerations in its design and development. During development and testing, Orbit will only use synthetic client data and will not process real financial or personal client information.
The production architecture is designed primarily around privacy principles, including minimizing the information provided to AI systems and applying PII masking before information is sent to AI models. Access to client information will also be restricted using role-based access control so that advisors can only access information they are authorized to view.
Due to Quebec's distinct privacy requirements, such as those introduced by Law 25, Orbit will be designed to comply with these laws. Data storage and processing will use Canadian data centres where applicable, reducing unnecessary exposure to foreign jurisdictions.
Orbit handles potentially sensitive financial-advisory information. That is why security must be considered in every layer of the application, rather than only at the database level.
The application will use HTTPS/TLS for data transmitted between components, secure password hashing such as bcrypt for authentication credentials, encryption for stored sensitive data, and role-based access control. Authentication and authorization mechanisms will be implemented so that users cannot access another advisor's clients or information without appropriate authorization.
Audit records will also be kept and maintained by the system for important operations to help identify unauthorized changes and provide traceability and logging when reviewing the system's behaviour. These include AI-generated content, advisor approvals, and changes to client information.
Orbit's AI features are meant to help advisors, not take the place of their expert opinions. As such, AI-generated summaries, messages, tasks, and follow-up recommendations may contain errors or unsupported information. That is why AI-generated content will not be sent to clients on its own. It will not be considered as information.
Instead, an advisor must review and approve AI-generated outputs before important actions are performed. This provides human oversight over the system and reduces the risk of incorrect information being communicated to clients which could otherwise be generated by the AI.
When possible, Orbit will provide source information for AI-generated recommendations and factual claims. This allows advisors to easily verify the information used by the AI before approving an action.
AI-generated information can be wrong, incomplete, outdated, or misleading. This is particularly important in a financial-advisory environment where incorrect information can lead a client to make a wrong choice with very bad consequences.
Orbit will therefore not let generative AI independently make financial decisions or provide financial advice directly to clients. Financial calculations and other important deterministic operations will instead use auditable software logic, rather than rely solely on generative AI's unreliable output.
The system will also evaluate AI functionality using measurable criteria such as conflict-detection accuracy and source-grounding accuracy. Claims that are not supported should be flagged instead of being displayed as true facts.
Client information and changes to client state must remain traceable. Orbit will keep records of important interactions and modifications, such as timestamps, the client involved, AI-generated drafts and proposed changes, as well as whether the advisor approved them.
The system's reconciliation mechanism will also avoid allowing AI agents to directly overwrite shared client information. Instead, agents will propose changes that can be evaluated against the existing state. Conflicting or potentially stale changes can then be flagged for review.
This approach helps preserve accountability and makes it possible to determine how important information was generated or modified.
It is imperative for clients to be informed when AI-assisted functionality is being used as part of their advisor's workflow. Rather than replace the advisor's relationship with the client, Orbit is simply intended to support the advisor's administrative work throughout the process.
Clients will remain able to communicate with their advisor, submit information, complete tasks, and access shared documents through the platform. Any content generated by AI that is meant for a client must be approved by the advisor before it is sent.
Orbit must ensure that automation does not reduce the level of responsibility placed on the financial advisor. Consequential decisions must always remain under human control, while administrative and organizational tasks can be deferred to the system's assistance.
By using multiple AI agents, it creates a risk that inconsistent or conflicting information could be introduced into a client's record. Orbit addresses this through its reconciliation mechanism, which detects conflicts, then duplicates and stale updates before they become finalized changes.
The overall design principle can be summarized as human-supervised automation. AI can propose, organize, summarize, and identify possible actions all it wants, but the advisor still remains responsible for having the final say in approving consequential changes and communications.