Every Closer reply, Maven post, Forge workflow and Sage outreach passes through Shield before send. Regulated-advice keywords, financial-promotion language, safeguarding flags, GDPR-special-category data, brand-tone drift — caught in 180ms, routed to a human if material, logged either way.
Four jobs. Every message. No exceptions.
Every outbound reply, post, email and DM is checked against a rules matrix in under 200ms. Regulated-advice, financial-promotion, safeguarding-disclosure and AI-disclosure rules all run inline.
Shield knows the difference between a CQC-regulated dental clinic, an FCA-authorised broker and an unregulated café. Different ruleset per sector, automatically applied based on the tenant profile.
Every flag, every hedge, every "routed to a human" decision is written to a 12-month tamper-evident log. Useful when the regulator asks. Useful when you ask.
Weekly summary to the practice owner: what was flagged, what was hedged, what escalated, and the trend across the last 12 weeks. No noise. Just the numbers that matter.
Five rule clusters. Plain English. One example per rule.
| Rule cluster | Example trigger | What Shield does |
|---|---|---|
| Regulated advice | "what dose should I take" / "can you advise on this investment" | Strips the answer, replaces with "I'm not able to advise — let me get someone qualified", routes to named human. |
| Safeguarding | "I'm being hurt" / disclosure language in childcare/education/healthcare tenant | Hard stop. Live escalation to DSL/duty officer. No AI follow-up. |
| Brand tone | Too much hype / unsupported claims / "guaranteed results" | Re-drafts with hedged language. Logs the original draft for the operator. |
| AI disclosure | First reply on a new conversation | Appends the standard "AI-assisted, human reviewed" line on regulated channels. |
| UK GDPR | Special-category data appearing in outbound (health, religion, etc.) | Blocks send. Logs incident. Notifies tenant DPO. |
No badges we haven't earned. Qualifiers stay in until the audits clear.
Personal data stored in UK/EU, encrypted at rest, retained per tenant policy. ICO registration: pending verification.
CQC, FCA, SRA, REC, Ofsted, RCVS, Construction Act, ICOBS — twelve sector packs maintained and version-controlled. Updated within 30 days of regulator change.
Every flag and hedge stored tamper-evident for twelve months. Exportable on demand. Includes the original draft, the flagged term, the route taken, and the human (if any) who reviewed.
Three steps. Under 200ms. Every message.
Shield reads the draft message + context (channel, sector, conversation history, sender role).
Runs the draft against the active rule pack for that tenant in under 200ms.
Send / re-draft / escalate to human / hard-block. Always logged.
Shield is one of six live agents on the Agent Networks platform.
The compliance pack covers DPIA, SIG-Lite, sector rule packs and the full audit-log spec. Procurement-ready.
Or call 0207 252 3333