Cetus AI Labs is our open research and innovation arm — exploring the frontier of AI governance, agent safety, and sovereign AI infrastructure for Australian enterprise and government.
Our research focuses on the practical, applied problems of deploying AI safely at enterprise and government scale.
Investigating how policy can be enforced at the point of AI execution — before a prompt reaches any model — without introducing unacceptable latency or brittleness.
ActiveExploring containment strategies, behavioural guardrails, and human oversight mechanisms for multi-agent systems operating in enterprise environments.
ActiveBuilding models for understanding and predicting token consumption at scale, and developing intelligent routing strategies to reduce cost without degrading output quality.
ActiveDesigning reference architectures for on-premise and private cloud AI deployments that meet Australian government data sovereignty and security requirements.
ActiveDeveloping standards and tooling for producing human-readable audit trails of AI decision-making that satisfy regulatory and compliance requirements.
ActiveTranslating high-level responsible AI principles into operational controls that can be enforced programmatically within enterprise AI infrastructure.
ActiveOur team publishes findings, frameworks, and perspectives on the evolving challenge of enterprise AI governance.
An examination of why existing pre-deployment safety measures are insufficient for governing AI in production enterprise environments, and a proposed architectural pattern for runtime enforcement.
A practical reference architecture for deploying large language models and AI agent systems within Australian government environments, addressing data residency, access control, and audit requirements.
An analysis of token consumption patterns across enterprise AI deployments, identifying the key drivers of cost overrun and evaluating mitigation strategies including intelligent routing and prompt compression.
We believe the foundational tooling for AI governance should be open and auditable. These projects are available to the community.
The open specification for the Songlines AI control protocol — defining the interface between enterprise applications and the Songlines Control runtime layer.
A domain-specific language for expressing AI governance policies in a human-readable, machine-enforceable format. Designed to bridge the gap between compliance teams and engineering.
A benchmark suite for evaluating the effectiveness of agent containment strategies across a range of enterprise-relevant scenarios and risk profiles.
A curated library of infrastructure patterns for sovereign AI deployments, covering on-premise LLM hosting, private vector stores, and air-gapped agent execution.
We collaborate with researchers, practitioners, and organisations working on the hard problems of enterprise AI governance.
Cetus AI Labs welcomes collaboration with Australian universities, government research bodies, and enterprise practitioners. Whether you're working on AI safety, governance frameworks, or sovereign infrastructure — we'd like to hear from you.
We also offer a structured research access programme for organisations that want early access to our governance tooling in exchange for real-world feedback and co-publication opportunities.
Get in touch with the Labs team →Follow our research, contribute to our open source projects, or speak with our team about how Songlines Control can work for your organisation.