ASD's new guidance treats AI as a threat, a security tool and an asset in its own right — and puts accountability for it with the board, not just IT.
In October, the Australian Signals Directorate (ASD) published AI considerations within the Essentials, part of the ASD Essentials series that is replacing the Essential Eight. It's the clearest statement yet from ASD on how AI fits into day-to-day cyber security practice, for any organisation that consumes, builds or deploys it.
Three ways to think about AI
ASD frames AI three ways at once. As a threat: it has accelerated existing attacks while introducing new attack surfaces, including synthetic voice and video used to impersonate staff to help desks or approvers, and "indirect prompt injection" hidden in emails or documents that AI tools process. As a solution: AI can triage alerts and speed up threat detection, but only if the AI tools themselves are governed like any other security product. And as an asset: every model, dataset, prompt log and AI-connected account is now something your business has to identify and manage.
ASD is also blunt about limits: AI can support decisions, but accountability for the outcome still sits with people. Outputs should be checked before they inform anything significant.
The board can't outsource this
ASD's Information Security Manual now includes an Executive AI accountability principle. It puts responsibility for AI being secure, controllable, human-supervised and used ethically directly with the board or executive committee, backed by actual processes for approval, risk assessment and ongoing review of AI use cases — not a policy statement that sits in a drawer. This is effectively a governance system for AI, and it lines up closely with what ISO/IEC 42001 asks organisations to build.
What counts as an "AI asset"
ASD's list is wider than most businesses expect: models, the infrastructure that hosts or trains them, external AI services, training data, prompts and outputs, the APIs that connect everything, and — easy to miss — the accounts AI agents use to act on your systems. ASD is explicit that these "non-person identities" need the same access governance as any staff login.
What to do now
Build an inventory of AI tools, models and AI agent accounts in use across the business, including ones staff adopted without asking IT.
Put an actual approval and review process behind AI use, owned at executive level, not left to whoever signed up for the tool. For ISO 27001 organisations, the asset side of this sits under A.5.9; the identity side under A.5.16.
Set a rule that AI-generated outputs get checked by a person before they drive a real decision, payment or access change.
Executive accountability for AI, an asset register that actually includes AI, and governed identities for AI agents are exactly what an AI management system is for. Ask Cybervisory today how we can help you build one, aligned to ISO 42001 and ready for what ASD now expects.
Key points
- ASD's new guidance treats AI as three things at once: a threat, a security capability, and an asset that must be inventoried and governed.
- A new ISM principle puts executive accountability for AI squarely with the board, not IT alone.
- AI agents can hold their own accounts and permissions — these need the same governance as any other access pathway.
Sources
Need help applying this to your organisation? Talk to Cybervisory.