AI Disclosure Statement

AI runs in two places. In the Plant Room workspace, drafts are produced for a named reviewer to approve before anything is issued. On our public website, the Proof critique at /prove, when it is switched on, prepares a first review of one document that a visitor chooses to upload: we check the file, extract its text within fixed limits (up to 100 PDF pages and 200,000 characters), and send that text - or, for a PDF with no text layer, the PDF itself - to Anthropic's Claude Sonnet 5 model. The critique is shown to the visitor, and emailed as a PDF only if they ask, without a BuiltAI person checking it first; it is labelled as needing the visitor's own review. The optional Explore service adviser is a separate, bounded feature. If it is made available, a visitor may send a free-text question to Anthropic, where a model generates a response based on approved BuiltAI service guidance. It is not an unrestricted chatbot. It cannot accept files, give legal, safety or investment advice, approve work, quote bespoke work, book meetings or submit enquiries. The Proof critique remains a separate, visitor-selected document-upload feature described above.

In the Plant Room workspace, AI outputs are labelled as drafts and require human review before they are issued. Workflow deliverables cannot bypass the QA1 and QA2 gates or, for client-facing documents, client approval. The public Proof critique is different: it is an automated first review that a BuiltAI person does not check before it is shown or emailed, so the visitor reviews it before relying on it.

Inputs to AI are classified Green, Amber or Red. Red-classified data is never processed by AI in the MVP. Amber data triggers a source-backed review requirement before output is used. See Governance for the full rules.

Procurement teams typically require a Data Protection Impact Assessment before deploying AI workflows. We publish a BuiltAI-specific DPIA starter at /governance/dpia covering engagement context, data flows, lawful basis, risks, mitigations, data subject rights and sign-off - sized for SME procurement, not a substitute for legal advice.

In the Plant Room workspace, every AI action - task type, input documents, classification, prompt template, model used, output reference and human review status - is logged for audit. Attempted Red-data processing is also logged.

• It does not provide legal advice.
• It does not certify engineering design.
• It does not provide statutory health and safety approval.
• It does not autonomously triage service desk tickets in production.
• It does not auto-issue proposals, SOWs, RAMS or board reports.