WitnessOS
Flinders

The definition comes first

Empire Labs runs two things. One defines what evidence has to exist. The other produces it.

WitnessOS by Empire Labs Pty Ltd

Two halves of one argument

A definition, then the product that meets it

Flinders is our engagement programme with the Australian bodies setting the rules for AI assurance: the Department of the Prime Minister and Cabinet's AI consultation, the South Australian Royal Commission into AI and the Australian AI Safety Institute. Its deliverable is a definition, not a product. Two questions. What counts as a reportable AI incident. And what evidence must exist to support one.

The programme is named after Matthew Flinders, who charted the coast of the continent he named Australia and left charts other navigators steered by for decades. Two things about that matter here. The chart was built from measured observation rather than assertion. It outlived him as the reference others used.

WitnessOS is the product. It is the implementation that produces evidence a third party can check without trusting us.

Revenue follows the definition. It never leads it.
Where they meet

One vocabulary, two jobs

Flinders contributes the vocabulary. Three parts of it matter.

The evidence ladder, which names how strong a record is rather than whether it exists at all.

The difference between a control that was declared and a control that was observed, which is the difference between a policy and a fact.

And the principle that a record should be checkable by a party who does not operate the system that produced it. That one is the whole argument in a sentence.

WitnessOS implements that vocabulary. Nothing here asks you to take our word for it. We are not claiming to be the standard, because the Commonwealth assigns standards status and not us. We are contributing a precise definition to public consultations and then meeting it with something that works.

The call

Start with the definition

If your organisation is preparing for the Australian cryptographic transition deadlines or needs to establish what evidence it holds for the AI systems it operates, we would like to hear from you.