AI Security
Adopt AI with confidence — governed, tested and secure.
Govern, test, adopt and protect
AI Governance and Risk
Govern AI use with policies, risk assessment and frameworks your board can stand behind.
AI Security Assessment
Test AI applications and integrations against AI-specific attacks, from prompt injection to data leakage.
Secure AI Adoption
Discover shadow AI, set guardrails for everyday AI use, and build with secure-by-design patterns.
AI Data Protection
Protect the personal and sensitive information flowing into prompts, models and training data.
How each capability works
AI Governance and Risk
AI governance in Australia is a moving target — guidance has already been replaced once, mandatory national standards have been announced, and existing laws apply in the meantime. We track that landscape so you don't have to, and build governance structured so tomorrow's mandatory standards land as an adjustment rather than a rebuild.
- AI use policy and acceptable-use rules for staff, systems and suppliers.
- AI governance framework aligned with the Guidance for AI Adoption's six essential practices — accountability, impact assessment, risk management, transparency, testing and monitoring, and human control.
- AI system inventory and register, with a named owner for every AI system in use.
- Use-case risk assessment using the NIST AI Risk Management Framework and its Generative AI Profile.
- ISO/IEC 42001 AI management system readiness — gap assessment, AIMS design and preparation for certification by an accredited certification body.
- Board and executive reporting on AI risk, and alignment with obligations that already reach AI systems through our Governance, Risk and Compliance service.
AI Security Assessment
AI applications fail in ways conventional testing doesn't look for. A web app scanner won't find a prompt injection that talks your chatbot into leaking customer records, and a code review won't catch an agent with more access than any employee would ever be given. We assess AI systems with people who understand both application security and AI attack classes.
- Threat modelling for AI systems: what the model can see, what it can do, and who can influence it.
- LLM application assessment against the OWASP Top 10 for LLM applications: prompt injection, sensitive information disclosure, insecure output handling and excessive agency.
- RAG and data pipeline security review: what your retrieval layer exposes, and to whom.
- AI supply chain review: third-party models, APIs, plugins and the permissions they arrive with.
- Agentic workflow risk assessment: tool access, delegation chains and human-in-the-loop controls.
- Guardrail effectiveness testing: whether your filters and controls actually hold under adversarial input.
Secure AI Adoption
Banning AI doesn't stop AI — it just stops you seeing it. Staff who are told no keep using personal accounts, and the data keeps leaving, invisibly. We take the enablement path: discover what's actually in use, sanction tools that can be governed, and set guardrails that make the safe way the easy way.
- Shadow-AI discovery — which AI tools are actually in use, by whom, with what data.
- Sanctioned tool selection and secure configuration — enterprise controls, data handling and retention settings.
- Guardrails for everyday AI use — what can and can't go into a prompt, enforced by controls rather than memos.
- Secure-by-design patterns for teams building with AI — least-privilege model access, output validation and human-in-the-loop checkpoints.
- AI usage training and awareness, delivered with our Cyber Security Training service.
- Ongoing review cadence, so new tools and features are assessed before they spread.
AI Data Protection
Most AI risk lands on data: personal information pasted into prompts, customer records swept into training sets, model outputs that reveal more than anyone intended. We assess AI data handling against the OAIC's guidance on privacy and generative AI, working hand in hand with our Privacy Advisory service — privacy decides what your organisation may do with personal information; this capability makes sure your AI systems actually do only that.
- AI data flow mapping — what enters prompts, training and fine-tuning, and where outputs go — assessed against the OAIC's guidance on privacy and generative AI.
- Vendor AI data handling review — whether your suppliers train on your data, and what your contracts actually say.
- De-identification and minimisation controls for AI pipelines, and readiness for incoming automated decision-making transparency requirements.