Model inventory & lineage
A written register of every model in production, every dataset behind it, every owner, and every external dependency. The starting point for any defensible governance posture — and the first thing a regulator asks for.
Governance, risk and optimisation for AI systems already in flight.
Drift monitoring, model inventories, access reviews, incident processes and cost optimisation — aligned to the AU AI Safety Standard and ISO/IEC 42001. Light-touch governance that operators will actually follow.
Dynamis Advisory — Governance provides decision-grade counsel on AI governance, risk and optimisation. Model inventories, lineage, drift monitoring (WhyLabs, Evidently, MLflow), access policy, incident response and cost optimisation. Aligned to the Australian Government Voluntary AI Safety Standard, ISO/IEC 42001, ISO/IEC 23894 and the NIST AI Risk Management Framework. Documentation prepared so external ISO certifying bodies (BSI, JAS-ANZ-accredited) can audit cleanly.
Workstreams
Where Governance counsel earns its place.
The work that keeps a model in production through audits, incidents, vendor changes and the inevitable cost spike. We write governance that engineers will read, not lawyers will hide.
Drift & performance monitoring
Statistical drift detectors, evaluation cadence, golden-set benchmarks and the alarm thresholds that wake an engineer up. The work that turns "the model is fine" from a feeling into a metric with a chart.
Access, policy & incident process
Who can deploy, who can retrain, who sees what data, and what happens when something goes sideways. Aligned to the AU AI Safety Standard and ISO/IEC 42001 — light enough that operators will actually follow it.
Cost & FinOps optimisation
Token budgets, GPU utilisation, model right-sizing, caching strategy, and the contract terms that stop a runaway bill in month seven. Governance and finance are the same conversation here.
Standards we align to
The references we work from.
We write to operators, not to certificate auditors. But we write so the certificate auditors can read it, too.
-
AU AI Safety Standard
The Australian government voluntary AI safety standard for risk-based controls.
-
ISO/IEC 42001
The international management-system standard for AI — the AI equivalent of ISO 27001.
-
ISO/IEC 23894
AI risk management guidance, applied to model lifecycle and deployment.
-
NIST AI RMF
Where US-aligned vendors and partners need a common reference, the NIST AI Risk Management Framework.
Engagement shape (Briefing, Review, Fractional) lives on the Advisory overview.
Common questions
FAQs
Here are some of our most frequently asked questions. Can't find what you're looking for? Reach out to our support team.
What is the AU AI Safety Standard?
The Australian Government Voluntary AI Safety Standard, published by the Department of Industry, Science and Resources in 2024. Ten guardrails for organisations developing or deploying AI: accountability, risk management, transparency, contestability, human oversight, fairness, safety, security, recordkeeping, and stakeholder engagement. Voluntary today, likely informative for regulated sectors as policy hardens.
How do you measure model drift in production?
Three layers: input drift (statistical tests on inference inputs vs the training distribution — Kolmogorov-Smirnov, population stability index), output drift (output distribution shift, calibration shift), and performance drift (accuracy decay against a held-out evaluation slice or a delayed ground-truth signal). Tools commonly recommended: WhyLabs, Evidently, MLflow, custom dashboards.
How do you control LLM cost without breaking the workload?
Token budgets per workload, model right-sizing (the smaller cheaper model often suffices for the routing layer), prompt caching, response caching, semantic deduplication of repeated queries, and a written FinOps contract that triggers an alert before the bill spikes. Cost optimisation is governance work, not procurement; it lives with the operating discipline.
Does Dynamis help with ISO/IEC 42001 certification?
We map governance work to ISO/IEC 42001 (AI Management System) so the documentation an external certifying body would request is already produced. Dynamis is not an accredited ISO certifying body — the certificate audit is conducted by the chosen certifier (BSI, JAS-ANZ-accredited bodies). We prepare the management system; they assess it.
What is the difference between governance and architecture?
Architecture decides how the system is built — pipelines, deployment topology, MLOps. Governance decides how the system is run — who can deploy, who sees what, what breaks the kill-switch, how an incident is handled. Architecture is the artefact; governance is the operating discipline. Both are needed; both are written down.
Start a conversation
One architect, one inbox.
Bring us the situation. We’ll pair you with a solution architect and write back — no hand-offs across divisions, no sales cadence.