1. Scope and permissible assistance
Enabled AI functions may help organise information, identify items for review or prepare a suggested analysis from authorised project records. They are assistive features, not an autonomous project manager, engineer, quantity surveyor, lawyer or safety assessor. The service does not delegate approval authority to a model. The firm must choose workflows appropriate to the information and the competence of the reviewer, and must not use AI to make solely automated decisions about people with legal or similarly significant effects without a separately assessed lawful arrangement.
2. Human approval and professional responsibility
Before acting on or releasing an AI result, an authorised person must inspect the source records, context, units, dates, quantities and relevant constraints, correct errors and follow the normal project approval workflow. Construction safety, certification, contractual notices, payment decisions and compliance determinations require the responsible qualified person’s judgement. A confident answer, structured JSON, a model label or a citation does not establish professional accuracy. Keep the reviewed source and material human changes with the decision record so others can understand the basis.
3. Uncertainty, sources and malicious inputs
Model results can contain omissions, invented details, outdated assumptions or inconsistent calculations. Citations must be checked against the actual source; missing evidence should remain a stated uncertainty rather than a completed finding. Documents and prompts may contain malicious instructions or misleading text, which must not override account permissions or approved workflow instructions. Users should report a suspected output or source-manipulation problem and avoid repeating an unreviewed action. AI output is not independent field evidence, a completed audit, a signed legal opinion or proof a provider action occurred.
4. Data scope and provider disclosure
OpenAI is the identified AI dependency, but account-specific processing location, retention/training settings, applicable terms and transfer safeguards remain unverified. The firm must authorise the data and workflow under its DPA and minimise prompts to what is needed. Do not include passwords, recovery keys, unneeded special-category data or material that cannot lawfully be supplied to the active provider. A provider name or encrypted storage backup does not establish private-model execution or verified EU / UK AI residency. Any additional model/provider path requires disclosure, qualification and applicable processing checks before customer use.
5. Allowances, eligibility and metering
Current Starter, Team and Portfolio plans include 25, 60 and 150 AI reviews respectively per monthly allowance period, subject to the accepted order and enabled workflow definitions. Availability depends on configuration and separate qualification; an allowance does not guarantee a particular model, unlimited retries or a feature that is not enabled. Capacity/metering controls may reserve an allowance for work in progress and retain an uncertain state until reconciled when provider completion is ambiguous. No automatic paid overage, measured cost-per-review or confirmed customer charge is claimed by this policy.
6. Review records, reporting and changes
The project team should preserve the relevant reviewed output, source references and human approval in its normal record workflow. Provider receipt and usage records can help explain processing, but do not alone prove output correctness, disclosure compliance or downstream delivery. Report issues through the support form or support@buildstance.com without sending secrets or unnecessary sensitive attachments. Material changes to providers, data use or decision authority require a revised notice and applicable agreement/consent. This draft remains subject to legal and operational review and does not remove the service’s commercial admission gates.
