Skip to main content
Assistant outputs are intended for decision support. They do not replace legal, tax, or regulated professional advice.

​
Core principles

Use non-alarming, actionable language focused on user next steps.
Ground responses in authorized data and active product context.
When data is missing or conflicting, state uncertainty clearly and avoid fabrication.
Respect tenant and user access boundaries for every generated response.
Do not provide regulated legal, tax, or financial advice; provide guidance and escalation paths instead.

​
v1.8.5 safety updates

v1.8.5 helps users understand which information Coach checked and how estimates differ from recorded facts.
  • Help answers distinguish supported, unsupported and unknown product behaviour.
  • Financial answers keep observed values, assumptions and projections separate.
  • Longer cash-planning horizons include explicit confidence and planning-use limitations.
  • Coach explains what information it could check and where gaps remain.
  • Public-source claims retain the applicable supporting evidence through the final answer.

​
v1.5.8 safety updates

v1.5.8 strengthens graceful degradation, spending explanations, and data-grounded coaching.
  • Coach gives clearer limitation messages when a tool, model call, context assembly step, or data source is unavailable.
  • Responses should avoid fabrication and offer a next-best action when complete data is not available.
  • Spending answers can use meaningful transaction categories where available, improving the evidence users see for financial explanations.
  • Sector context, behavioural signals, and research into external support opportunities can provide richer grounding while preserving the boundary between guidance and regulated advice.
  • Eligible conversation analytics can support quality review through topic, intent, sentiment, and confidence signals.

​
Safety controls in practice

Coach keeps known financial facts separate from decision uncertainty. An available bank balance can inform a spending discussion without establishing that all of it is freely spendable. Missing commitments, reserves or expected returns are identified as specific limitations. Relevant conversations apply credential protection, account-action boundaries and professional-advice limitations. Configured partner tone does not override those boundaries or the evidence supporting an answer.

Prompt + routing guardrails

Detect high-risk or out-of-scope requests and route to restricted handling paths.

Runtime disclaimers

Show explicit user-facing disclaimers for known uncertainty or processing states.

Refusal behaviour

Refuse disallowed requests with clear rationale and safe alternatives.

Validation suites

Test tone, scope, and trigger behaviour to catch regressions before release.

​
Escalation model

1

State limitation

Explain confidence gaps or missing context explicitly.
2

Constrain recommendation

Avoid prescriptive advice in regulated or high-risk scenarios.
3

Escalate to human channel

Route users to approved human support workflows when needed.