Will professionals rely on AI output for regulated advice?
AI may support research, issue spotting and drafting, but a competent professional needs to verify material sources and own the final regulated advice. Record checks proportionately without duplicating confidential inputs unnecessarily.
A fluent draft can still misstate a source or omit a client fact. Before using it in advice, the reviewer needs to be able to check the material claims, question the assumptions and correct the draft.
The practical answer is assisted judgement, not delegated advice
Set permission for the specific research, analysis or drafting task. Keep the final advice decision with a competent professional under the engagement's sign-off rules. Permission to prepare options does not authorise an unreviewed recommendation or client message.
- Allowed: summarising background documents, extracting issues, preparing a first-pass checklist, comparing clauses, drafting options and flagging inconsistencies.
- Conditional: risk scoring, recommendation drafts and client-specific conclusions, where a qualified reviewer validates the inputs, reasoning and limitations.
- Not acceptable without stronger controls: automated advice, unreviewed client communications, confidential-data uploads to unapproved tools, or unexplained outputs used as evidence.
Check where AI can safely speed up regulated workflows
Why regulated advice needs a higher bar than ordinary AI use
Identify the requirements that apply to the profession, jurisdiction and service, including competence, suitability, confidentiality, supervision and record keeping where relevant. Assess the AI-assisted task against those requirements rather than assume the same rules apply to every type of professional work.
Check whether the reviewer can explain the sources, changed assumptions and reasons for accepting the advice. A plausible answer without those checks should remain a draft, even if its wording needs little editing.
A complaint or review may require the firm to explain its reasoning. Choose records that support that explanation, subject to the engagement's access and retention requirements.
The controls that make AI-assisted advice defensible
A professional services firm should make reliance conditional on a few practical controls rather than a vague AI policy.
- Approved tools: define which AI systems can be used for client work and which are banned.
- Data rules: state what client, personal, privileged or commercially sensitive data may be entered into each tool.
- Human review: require a named reviewer for client-specific advice and high-risk recommendations.
- Source checks: make reviewers verify legal, financial, technical or regulatory claims against trusted sources.
- Version records: retain the review and decision evidence needed for the advice, with source references where sufficient. Keep full prompts, source copies or outputs only where justified and permitted by confidentiality, access and retention requirements.
- Escalation: define when AI-assisted work needs partner, compliance or risk review before release.
Build a lightweight AI governance operating model
How to decide what level of reliance is acceptable
Not every use case needs the same control burden. Segment AI work by client impact and reversibility.
- Low risk: internal brainstorming, formatting, generic summaries and non-client-specific research notes.
- Medium risk: first drafts, due diligence summaries, policy comparisons and risk registers that feed into professional work.
- High risk: advice recommendations, suitability judgements, legal interpretations, coverage positions, regulatory filings and client-facing conclusions.
For high-risk uses, the firm should be able to show the evidence trail and the human reasoning that turned an AI suggestion into professional advice. If that cannot be shown, the output should remain internal working material.
What leaders should document before scaling use
Before encouraging wider adoption, leadership should document the firm’s position on four points: permitted tools, permitted data, review standards and retention requirements. Keep this short enough that people will use it.
Test those rules against a task the team already performs. The reviewer should be able to identify the permitted input, checks and escalation route before relying on the output.
Implement AI workflows with review and audit trails built in
Conclusion
Use AI-assisted material only to the extent the professional can justify it against the client facts and sources. Keep ownership and escalation clear, and retain proportionate evidence of the checks. Faster drafting does not establish that advice is suitable.
Need a safer AI route for your firm?
If an AI draft is being used in advice, map the source checks and professional sign-off for that task. Resolve any gap in permitted inputs, review or escalation before relying on it.
