Agentaus: Australia’s first Sovereign AI Model

For thousands of qualified Australian professionals, hesitation around artificial intelligence has little to do with a fear of technology. Lawyers in Melbourne, tax advisers in Sydney, and medical specialists in Brisbane understand that modern software can speed up routine drafting and document review. What holds them back is a far more fundamental obligation: protecting client confidentiality, legal professional privilege, and statutory privacy duties.

Until recently, testing generative AI meant sending sensitive work notes through overseas servers operated by foreign providers, often under complex terms of service. On 10 November 2026, Canberra-based AI developer Trellis Data changed that equation by launching Agentaus 1, a 220-billion-parameter sovereign AI model built specifically for Australian organisations. For cautious professionals who have been waiting on the sidelines, this development removes the largest privacy roadblocks standing in the way of a safe first step.

Professionals on the Sidelines: Client Confidentiality

In regulated Australian professions, trust is the foundation of every client relationship. Under the Australian Privacy Principles, practice leaders must take reasonable steps to protect personal and commercial information from unauthorised disclosure or cross-border exposure. Similarly, legal practitioners and registered tax agents face strict professional conduct rules governing how client files are stored, accessed, and processed.

When free consumer AI tools first surged in popularity, many firms rightly placed blanket restrictions on staff usage. Pasting a draft merger agreement, a patient referral letter, or a family trust restructure into an unvetted public website risked breaching client confidence and exposing sensitive records to external model training. Without a trusted local alternative, curious professionals were left with an uncomfortable choice between staying completely manual or risking compliance.

What Sovereign AI Means

The phrase sovereign AI is frequently used in policy briefings, yet its practical meaning for an everyday professional practice is straightforward. A sovereign AI system is developed, hosted, and governed within Australian legal jurisdiction, ensuring that sensitive records never leave the country.

Trellis Data, founded in Canberra in 2016 and backed by CSIRO-founded venture firm Main Sequence, built Agentaus 1 around this exact requirement. Through its Protected Mode, all models, prompts, encrypted documents, and generated outputs remain inside Australian data centres used by Federal Government agencies. Just as importantly, Trellis Data enforces a strict zero-training guarantee: client material stays private to each user and is never viewed, commercialised, or used to train shared models.

How Onshore Hosting Protects Regulated Australian Practices: Consider a mid-tier commercial law firm in Melbourne reviewing supplier agreements for a local healthcare client. Using an offshore consumer tool creates uncertainty around foreign subpoenas and cross-border data residency. By working inside a sovereign Australian workspace, the firm maintains clear legal jurisdiction under Australian law while giving solicitors a practical tool to summarise clauses and spot inconsistencies.

Inside Agentaus 1: Capability Built for High-Trust Work

Many Australian leaders have previously assumed that choosing a local provider meant accepting second-rate performance. Agentaus 1 challenges that assumption. At 220 billion parameters, the model is roughly 50 per cent larger than Trellis Data's earlier release, delivering strong reasoning and high accuracy on complex professional tasks while running efficiently on smaller compute clusters to keep ongoing operating costs predictable.

Security sits at the core of the platform's design. Trellis Data holds ISO/IEC 27001:2022 certification for information security management, reflecting nearly a decade of building systems for Australian defence, intelligence, and law enforcement agencies. Whether an organisation starts on a standard sovereign cloud plan or eventually requires on-premises or air-gapped deployment, that security pedigree translates into dependable safeguards without requiring an internal cyber security department to configure them.

Traceable Outputs and Source Verification: Accuracy matters just as much as privacy when signing off on professional advice. Agentaus 1 is built specifically for structured document workflows, enabling users to convert lengthy policy submissions, regulatory guidelines, or tender specifications into clear executive briefs with verifiable references back to original source files. A civil engineering consultancy in Brisbane or a public policy practice in Canberra can check every claim against its underlying document before sharing a draft with a client.

Practical Everyday Use Cases Across Australian Professions

You do not need complex technical integrations or custom software development to benefit from sovereign AI. Qualified professionals across Australia can apply onshore tools to contained, everyday administrative tasks where human judgement remains firmly in control:

  1. Accounting and tax advisory in Sydney: Summarising multi-year client correspondence and financial background notes into structured preparation checklists while keeping sensitive tax file numbers and entity structures inside Australia.

  2. Allied health and clinical practices in regional Victoria: Drafting clear referral summaries and patient care plans from consultation notes inside an encrypted local environment aligned with Australian health privacy standards.

  3. Commercial legal and compliance teams in Perth: Comparing incoming vendor contracts against an approved internal playbook to highlight non-standard liability clauses before a senior lawyer conducts final review.

Four Simple Steps to Start Safely with Sovereign AI

If you have been waiting for a trustworthy way to test AI in your practice, the arrival of home-grown sovereign capability makes it much easier to begin calmly. Follow four practical steps to keep your first trial disciplined and secure:

  1. Confirm data residency and non-training terms first: Before uploading any internal file, verify that Protected Mode or an equivalent Australian sovereign hosting tier is active and that customer inputs are excluded from model training.

  2. Start with a small, ring-fenced document set: Test the platform on non-sensitive internal templates, public regulatory guides, or de-identified sample files while your team learns how to write clear prompts.

  3. Keep a qualified professional in the approval loop: Treat every AI-generated summary or draft as preliminary groundwork, ensuring a trained practitioner verifies citations and professional accuracy before any document leaves the office.

  4. Give staff an approved local workspace: Publish a brief, one-page usage policy directing team members to your approved sovereign platform so confidential client records never drift into unmonitored consumer accounts.

Confidence Without Compromising Professional Standards

Qualified Australian professionals were right to be cautious during the early wave of public AI tools. Protecting client confidentiality and maintaining public trust are non-negotiable duties in law, finance, healthcare, and engineering. With Canberra-built models like Agentaus 1 proving that near-frontier reasoning can live entirely inside Australian borders under strict non-training guarantees, newcomers no longer must trade privacy for productivity.

Start with one contained document workflow, keep human oversight at the centre of your process and build confidence at a pace that suits your practice.

Next
Next

The Medicare Breach: Australia’s Catalyst for legislated AI Governance