§ About  —  Who we are

An Australian AI lab, chasing difficulty.

Glassbox Labs is an Australian AI development company. We were founded to take the best of what frontier AI research has produced and make it safe, legible, and commercially useful for the institutions that keep the country running.

Founded
2026 in Australia.
Team
Small and senior. ML research, production engineering, regulated-industry delivery, public policy.
Stance
We do not chase headcount. We chase difficulty.
Jurisdiction
Australian legal entity, Australian staff on record, Australian law.
§ 01   The name

Why "Glassbox".

The phrase "black box" is the honest description of how most modern AI systems are built and bought. Inputs go in, outputs come out, and very few people in the decision loop can say with precision what happened in between.

That is an acceptable state of affairs for a novelty. It is not acceptable for systems that price loans, triage patients, grade students, or advise ministers.

A glass box is the same object, differently designed. The mechanism is visible. The reasoning is attributable. The behaviour under stress is known, because it has been tested.

Our name is a commitment, not a metaphor.

§ 02   What we believe

Five principles that shape the work.

2.3.1Intelligence is an engineering discipline.Modelling, evaluation, deployment, and operations are one continuous craft. Treating them as separate functions is the most common reason AI projects fail.
2.3.2Transparency is a feature, not a tax.Making a system legible to its operators, auditors, and subjects is productive work. It surfaces defects earlier, reduces regulatory risk, and produces better systems.
2.3.3Sovereignty matters.Where a model is trained, where its weights sit, who can access its logs, and whose law applies when something goes wrong are first-order questions.
2.3.4Scale with restraint.We invest heavily in evaluation, red-teaming, and deployment discipline so that the systems we build get bigger only when they are ready.
2.3.5Write things down.We document our methods, publish our non-proprietary findings, and favour plain English over private jargon. If we cannot explain a decision to a non-specialist, we do not understand it yet.

If this reads like your kind of company,
come and find us.

Start a conversation → See what we do →