Hosting sovereignty
Domestic infrastructure, compute, data-centre and cloud capacity. This matters, but it is not the whole question.
Australian AI Fair-Go
Fit the tool to the job.
Australian AI policy should protect people from harmful deployments without taking away lawful user choice, local control, model plurality, accessibility support, and grassroots technical participation.
Operating principle
Give each system only what its job requires.
Fit the tool to the job. Give it only what that job requires. Keep the human in charge.
A school learning assistant, a local disability-support workflow, a small-business tool, a research rig, and a frontier cloud system do not need the same model, host, permissions, connectivity or regulatory treatment.
Policy wedge
Do not concentrate permission in order to control concentration.
Do not solve concentrated technological power by concentrating technological permission.
Rules designed to restrain powerful platforms can accidentally make those same platforms the only practical route to lawful AI participation. Australia should regulate harmful conduct, high-impact deployment and concentrated power without forcing every problem into one approved provider or architecture.
Three sovereignties
National capacity has more than one layer.
Institutional sovereignty
Australian agencies, universities, recognised companies and standards bodies can research, procure, deploy and govern AI systems.
Broad national capability
Australians across the country can learn, inspect, run, adapt, evaluate and build useful AI systems without first passing through an institutional gate.
Evidence spine
FMI can show working alternatives, not only critique.
Hybrid choice
Chatty-Cog and Chatty-mini demonstrate local-first systems with deliberate BYO cloud lanes where cloud capability is useful.
Proportionate restraint
Chatty-EDU shows local-only design as appropriate restraint for schools, children, privacy and teacher oversight.
Task-fit evaluation
Cognition Mesh Test Chamber evaluates model, host and task suitability instead of treating AI as one undifferentiated category.
Builder literacy
Nanochat LLM Tweaker, Janet School and LLM Semantic Dataset Sorter support practical learning, experimentation and data work outside institutional umbrellas.
Nanochat LLM Tweaker, Janet School, LLM Semantic Dataset Sorter
Human authority
AI Teaming Framework and the Chatty tools demonstrate explicit human direction, bounded workflows and model-neutral collaboration patterns.
Unmet ecosystem infrastructure
Platform-level user-authorised assistive delegation remains a policy and standards objective. FMI demonstrates adjacent architecture inside user-owned hosts; the external circuit is the missing piece.
Current official context
This is being written while Australian AI policy is actively forming.
National AI direction
Australian Government policy now includes the National AI Plan, the Office of AI, and Australia's AI Safety Institute.
Australian Government AI policy National AI Plan Office of AI AI Safety Institute
Live inquiry path
The Senate inquiry into artificial intelligence and data centres is accepting submissions, creating a direct path for evidence-led public input.
Legal and infrastructure context
Existing government material already connects AI adoption with legal duties, privacy, automated decision-making, data minimisation and infrastructure expectations.
AI adoption legal landscape Data-centre and AI infrastructure expectations
Core ask