Home · Platform
01 · Evidence platform

Make every AI governance
claim traceable.

The Mycelium Evidence Platform is the working layer behind the Mycelium 14-Dimension Framework. Every board-level claim — every met, partial or gap — traces back to a human-approved artefact, a named owner, a date and a decision. Built for Australian regulated organisations that can no longer afford to attest on faith.

02 · The problem

AI governance has scaled.
Evidence has not.

Boards are signing AI attestations they cannot defend. Risk teams produce slide decks; internal audit asks for trace. Spreadsheets and policy folders fall apart on the third quarterly cycle. Enterprise GRC platforms store documents but cannot prove a human reviewed each AI control on each cycle.

APRA CPS 230 and CPS 234, AS ISO/IEC 42001 and the ASD ISM AI controls now ask for evidence — and the AICD has put AI directly on the directors’ duty register.

Most platforms cannot do trace. Most consultancies cannot ship product. The Mycelium Evidence Platform is built to sit between them.

03 · How it works

Four phases. One traceable record.

01 · Capture

Live AI register

Every AI system registered, owned and risk-rated.

02 · Assess

Controls mapped

Each system mapped to APRA, ISM, AS ISO/IEC 42001, NIST AI RMF, AICD and the Privacy Act ADM reforms.

03 · Evidence

Human-approved

Evidence requested from named owners. Each artefact reviewed by a named human. Stamped with an evidence_uid.

04 · Publish

Auditor-followable pack

Drafted under validation, rendered to PDF, version-stamped, uploaded to the private client portal, audit row written.

04 · Capabilities

What the platform produces.

CAP · 01

Live AI register

Every system, owner, business unit, lifecycle stage and data sensitivity.

CAP · 02

Risk rating

Per-system rating with documented rationale, refreshed each cycle.

CAP · 03

Control mapping

Each system mapped to the 14 dimensions and to the regulator and standards set relevant to the organisation.

CAP · 04

Evidence requests

Plain-English asks issued to named owners. Tracked. Reminded. Logged.

CAP · 05

Human approval gate

Every evidence item approved by a named human reviewer. Appended, not overwritable.

CAP · 06

RAG scorecard

Met, partial, gap — per pillar, per system, per cycle.

CAP · 07

Evidence pack

Versioned PDF: executive summary, scorecard, register, control-evidence detail, gaps and remediation, methodology, attestation page.

CAP · 08

Gaps register

Named remediation owner and due date for every gap or partial.

CAP · 09

Client portal

Passwordless access, RLS-isolated client data, 120-second signed download links, append-only access audit.

05 · Security · traceability

The boundary is in the product.

Authentication
Passwordless magic-link via Supabase. Every session expires.
Authorisation
Postgres row-level security. Cross-client access returns HTTP 403 — never 404 — so pack IDs cannot be probed.
Private storage
Evidence packs in a private Supabase bucket. Access only via short-lived signed URLs minted by a server-side function.
Signed URL expiry
120 seconds by default. Tunable per client.
Append-only audit
Every publish and every pack view written to an append-only audit_log row. No update or delete grants on the table.
Versioned packs
Every published pack carries a status date and pack ID. Older versions remain queryable.
Fail-closed model
Claude Opus drafts under JSON validation with a Sonnet fallback. Malformed output is rejected before the pack is published.
Human review
Mandatory on every evidence item. The gate is enforced in the data model.
06 · Assurance boundary

You attest. We evidence.

Stated boundary

The Mycelium Evidence Platform reports what human-reviewed evidence demonstrates. It does not assert that an organisation is compliant, certified or assured. It does not replace management, board, legal, risk, audit or regulatory accountability. Compliance, certification and assurance opinions rest entirely with management, the board and the relevant assurance providers.

07 · Use cases

Where the platform earns its place.

USE · 01AI governance attestation cycle for boards and audit committees.
USE · 02Pre-audit evidence preparation — internal or external.
USE · 03APRA CPS 230 / CPS 234 alignment evidence.
USE · 04AI vendor and third-party assurance.
USE · 05Director-duty AI risk reporting under AICD guidance.
08 · Industries

Built for Australian regulated organisations.

IND · 01Banking, mutuals, insurance and superannuation under APRA.
IND · 02Energy distributors, gentailers and critical-infrastructure operators under the SoCI Act.
IND · 03Legal and professional services with internal AI capability.
IND · 04Federal and state public-sector AI programmes.
09 · Pilot offer

The AI Governance Evidence Pilot.

Six weeks. One AI system. All 14 dimensions. One published, board- and auditor-ready evidence pack. Design-partner pricing available for the first two to three clients.

Designed to fit inside an existing audit-prep or AI project budget.

Book a 30-minute conversation
AI GOVERNANCE EVIDENCE PILOT

1 system · 14 dimensions · 1 pack

Capture → Assess → Evidence → Publish. The full pipeline, end-to-end, in six weeks. Sponsor-signed. Auditor-followable.

SCOPE LOCKED · DECISION POINT AT WEEK 6
10 · FAQ

The questions buyers ask first.

Does the platform certify compliance?

No. The platform produces an evidence trail; assurance opinions remain with management, the board and the relevant assurance providers.

Where does the data live?

Sensitive artefacts remain in the client’s Microsoft 365 / SharePoint tenant. Mycelium holds references and metadata only. Evidence packs are stored in a private Supabase bucket, default region AU-Sydney.

Who can see another client’s data?

No-one. Postgres row-level security isolates client data. Cross-client access returns HTTP 403 and is written to the append-only audit log.

How does authentication work?

Passwordless magic link via Supabase. The service-role key never reaches the browser; download links are signed by a server-side Netlify function and expire in 120 seconds by default.

Does Claude decide compliance?

No. Claude drafts the pack narrative under JSON validation, with a fallback model and explicit fail-closed behaviour. Every claim is cited to an evidence item approved by a named human reviewer.

Can I see what a published evidence pack looks like?

Yes. There is an anonymised sample on this page (See a sample pack, top of hero), built to the same structure as a live cycle: executive summary, RAG scorecard, control-evidence detail with evidence_uid, gaps and remediation, methodology and an attestation page. All entity names, identifiers and email addresses are illustrative.

Can the platform integrate with our GRC?

Yes, via documented export. Custom integrations are available as a paid extension after pilot.

Is the platform IRAP- or SOC 2-certified?

Not yet. The platform is engineered to the relevant baselines; formal certification will be pursued when a contracted client requires it. The architecture, controls and limits are documented in the Security and Trust Summary, available on request.

11 · Start the conversation

If your board is being asked to attest to AI governance — or your auditor is.

The Mycelium Evidence Platform is the system that makes that attestation defensible.

Book a private demonstration
30-MINUTE DEMONSTRATION · VIDEO OR MELBOURNE · ADMIN@THEMYCELIUMGROUP.COM.AU
Or open the full 14-Dimension framework ↗