Guides> AI and ethics
AI and your ethical obligations as a CSF: what you need to know in 2026
Artificial intelligence is making its way into the practice of financial security advisors (CSFs): research, drafting, calculations, meeting preparation. Used well, it saves hours. Used poorly, it exposes the advisor to factual errors, confidentiality breaches and, ultimately, to their own professional liability. This guide covers what a CSF needs to know before bringing AI into their practice.
The basic principle: responsibility cannot be delegated
No tool, AI or otherwise, changes the framework in which you practise. Your code of ethics requires you, among other things, to act with competence and professionalism, to give the client accurate and complete information, and to protect the personal information entrusted to you. If incorrect information reaches the client, it does not matter whether it came from a search engine, a colleague or an AI: it is your professional signature on the line.
The right way to see AI: a very fast research assistant whose work must be checked, exactly as you would review an intern's draft before sending it to a client.
The three concrete risks of general-purpose tools
1. Hallucinations. Language models produce plausible answers, not necessarily true ones. In a numbers-driven field like ours, this shows up as invented limits, approximate thresholds or law articles that do not exist, all delivered with perfect confidence.
2. Outdated data.YMPE, TFSA and RRSP limits, tax brackets, the OAS recovery (clawback) threshold: these figures change every year. A model trained on last year's data will answer with last year's figures, without telling you.
3. Confidentiality.Pasting a client's full profile into a consumer tool means disclosing personal information to a third party. Quebec's Law 25 (Loi 25) has significantly tightened the obligations of Quebec businesses in this area. The reflex to develop: always anonymize.
Best practices for professional use
- Verify before you communicate. Any figure or rule intended for a client must be validated against an official source (CRA, Retraite Québec, Revenu Québec) or a tool that cites its own.
- Anonymize your questions.“55-year-old client, RREGOP, $95,000 salary” gives an answer just as useful as the full file, without a name, SIN, policy number or contract number.
- Demand sources. An answer without a verifiable source is a hypothesis, not information. Favour tools that cite their references.
- Document your process. Keeping a record of the source you verified protects your file if a recommendation is ever challenged.
- Keep the final say. AI proposes, the professional decides. No recommendation should leave your office without going through your judgment.
General-purpose or specialized: choosing the right tool
The question is not “AI or no AI”, but “which tool for which use”. A general-purpose tool is fine for rewording an email. For substantive questions (Quebec taxation, pension plans, compliance), a tool specialized for Quebec, which cites its sources and keeps its figures up to date, considerably reduces the verification work. That is precisely the philosophy behind Atlas CSF+: verified 2026 figures, cited sources, and safeguards designed for a CSF's practice, without ever claiming to replace your judgment.
Frequently asked questions
Can a financial security advisor use AI in their practice?
Yes, nothing prohibits using AI tools as an aid for research or drafting. But the advisor remains fully responsible for the accuracy of the information given to clients and for the compliance of their recommendations. AI is a reference tool, never a substitute for professional judgment.
What are the main risks of using ChatGPT for client questions?
Three main risks: (1) hallucinations, meaning plausible but false answers, including invented figures and law articles; (2) outdated data, since limits, thresholds and tax brackets change every year; (3) confidentiality, since pasting a client's personal information into a consumer tool may breach your obligations to protect personal information.
Can I enter a client's personal information into an AI tool?
Caution is required. Law 25 governs the collection, use and disclosure of personal information in Quebec. Best practice: never enter full identifiers (full name, SIN, policy or contract numbers) into an AI tool, and favour anonymized scenarios ("55-year-old client, RREGOP, $95,000 salary").
Who is responsible if the AI gives a wrong answer?
You are. Professional liability cannot be delegated to software. If a client suffers harm because of incorrect information you gave them, the fact that it came from an AI does not release you from responsibility. That is why checking sources before anything is communicated to the client is the golden rule.
What is the difference between a general-purpose AI tool and a specialized one?
A general-purpose tool (ChatGPT, Gemini) answers everything, with no guarantee on the figures or on the law that applies in Quebec. A specialized tool is built on the Quebec framework, cites its sources and keeps its figures up to date, which reduces (without eliminating) the verification work. In both cases, the professional judgment remains yours.
This guide is provided for information purposes only and does not constitute legal or ethics advice. For any question about your specific obligations, consult your code of ethics, your firm or a legal advisor.