AIHA Office Agentic AI & Hyperautomation Implementation Office · demo Acting as

Strategy & governance

Guardrails catalogue

Sixteen input, output, action, runtime and data guardrails with enforcement point and the risk tiers they apply to.

Guardrails

GR-01 Prompt-injection & jailbreak filter
GR-02 PII detection & masking (DLP)
GR-03 Data-class routing (restricted data never leaves the enclave)
GR-04 Grounding check ≥ 0.90 (cited sources)
GR-05 Independent critic model (dual-model check)
GR-06 Deterministic validators (regex, JSON schema, business rules)
GR-07 Toxicity, bias & tone filter for customer text
GR-08 Tool allow-list per agent (MCP registry)
GR-09 Financial limit per action (e.g. € 10 000 payment release)
GR-10 Four-eyes rule on irreversible actions
GR-11 Sanctions / PEP hit always to expert
GR-12 Step, token and cost budget per run
GR-13 Loop & runaway detection, kill switch
GR-14 KPI-triggered pipeline lock (CSAT < 85%, NPS < +30, CES > 3.5)
GR-15 Retrieval access control (user rights applied to RAG)
GR-16 Immutable audit trail of prompts, tools and decisions

Public demo with fictional data (Example Group). © 2025–2026 Russo-Got Marius Mihail & PIR2-IT. All rights reserved.

More in this demo