Skip to content
AG.00AI governance

Public AIBOM, ahead of EU AI Act enforcement.

AI Bill of Materials per module, model cards per surface, EU AI Act risk-tier classification, red-team summaries, C2PA content provenance. Most platform vendors will get to this in 2026 because they have to. FlyttGo is here in 2025 because procurement asks.

  • AI surfaces · live
    8
  • High-risk · EU AI Act
    3
  • Surfaces with HITL
    5
  • Unacceptable-risk surfaces
    0
AG.PRGovernance pillars

Four anchors for AI transparency.

  • AG.PR.01

    AIBOM · AI Bill of Materials

    Per-module declaration of every model in production: weights origin, training data class, retraining cadence, drift telemetry. Refreshed in the changelog every release.

  • AG.PR.02

    Model cards on every surface

    Each model surface ships a model card declaring intended use, out-of-scope use, performance bounds, fairness evaluation, demographic and geographic coverage, known failure modes.

  • AG.PR.03

    EU AI Act risk-tier classification

    Pre-emptive classification of every AI surface against the EU AI Act risk taxonomy (minimal / limited / high / unacceptable). Public registry, refreshed when a surface changes risk-tier.

  • AG.PR.04

    Red-team summary archive

    Annual external red-team exercises against every high-risk surface (Identra liveness, Payvera fraud, Civitas triage). Executive summaries published; technical reports available under MNDA.

AG.ABAIBOM · live registry

Eight surfaces. Every model accounted for.

CodeModuleSurface · model · weightsRisk-tierHITL
AG.S.01Transify
Provider routing optimiser
Routes orders across providers optimising cost × time × coverage under regulatory constraints (cabotage, hazmat).
model: Custom multi-objective optimiser + tabular regressionweights: In-house · weekly retrained
LimitedNo
AG.S.02Workverge
Shift dispatch optimiser
Suggests shift coverage and dispatch order under labour-law constraints. Operator approves the slate.
model: Constraint solver + heuristic warm-startweights: In-house · daily retrained
LimitedYes
AG.S.03Civitas
Application triage classifier
Triages incoming citizen applications into the responsible agency. Civil servant validates before action.
model: Fine-tuned multilingual transformerweights: In-house · monthly retrained
HighYes
AG.S.04Identra
Liveness + spoofing detection
Liveness detection during identity verification. Adversarial-tested against PA-DSS 4 attack vectors.
model: Vision transformer · liveness ensembleweights: Hybrid · vendor-licensed core
HighYes
AG.S.05Payvera
Anomaly + fraud scoring
Real-time anomaly scoring on payment intents. Falls back to rule engine if model uncertainty exceeds threshold.
model: Gradient-boosted ensemble + behavioural sequence modelweights: In-house · streaming-updated
HighYes
AG.S.06EduPro
Cohort analytics aggregator
Privacy-preserving cohort analytics — no per-student inferences ever leave the boundary.
model: Differential-privacy-aggregated tabular modelweights: In-house · DP-SGD trained
LimitedNo
AG.S.07AskFlyttGo
Procurement assistant (planned)
Drafts CAIQ responses, RFP responses, custom proposals. Every artefact carries a signed provenance manifest.
model: Anthropic Claude · Sonnet tier · grounded retrievalweights: Vendor · context-only
LimitedYes
AG.S.08FlyttGo Marketplace
Pricing intelligence
Per-region reference pricing surfaced to buyers; anomaly flags on under/over-priced supply for the operator.
model: Per-corridor reference price + anomaly detectionweights: In-house · daily refreshed
LimitedNo

Risk-tier follows the EU AI Act taxonomy. Surfaces categorised ‘high’ carry mandatory HITL, full model card, fairness evaluation and quarterly drift report.

AG.RMCompliance timeline

Six quarters tracked publicly, ahead of enforcement.

  1. AG.RM.Q3.25Q3 2025

    Initial AIBOM published

    First public AIBOM listing every AI surface across the platform. Voluntary; not yet a regulatory requirement.

  2. AG.RM.Q4.25Q4 2025

    Model cards for high-risk surfaces

    Identra liveness, Payvera fraud and Civitas triage shipped first model cards.

  3. AG.RM.Q1.26Q1 2026

    Risk-tier classifications complete

    Every AI surface tagged against the EU AI Act risk taxonomy. Registry refreshed monthly.

  4. AG.RM.Q2.26Q2 2026

    External red-team baseline

    First external red-team exercise across all high-risk surfaces. Executive summary published.

  5. AG.RM.Q3.26Q3 2026

    EU AI Act enforcement begins

    High-risk obligations enforceable from August 2026. FlyttGo positions ahead of the line, not at it.

  6. AG.RM.Q4.26Q4 2026

    C2PA content provenance · GA

    Every AI-generated artefact (proposals, summaries, CAIQ responses) signed with a C2PA provenance manifest.

AG.NXWhere AI governance plugs in

AI transparency is one layer of the trust framework.

The AIBOM doesn't stand alone — it sits inside the broader trust posture and the agent surface. The four pathways below take a programme from this manifest into a signed engagement.