The Australian Government’s newly released PSPF Policy Advisory 001-2026 (26/05/2026), Cyber Security Readiness in the Frontier AI Era, represents a significant step forward for national cybersecurity guidance. The advisory recognises a reality that many security leaders are now wrestling with every day. Artificial intelligence is no longer merely a productivity tool for niche verticals and is rapidly becoming an autonomous participant in almost every business process, executive decision, and daily system operation. Indeed, as AI transitions from personal assistant to business identities, traditional cybersecurity controls are too insufficient to manage risk. Organisations must adapt their security strategies to account for a new category of privileged identity: the AI agent.
What makes this advisory particularly important is its emphasis on readiness rather than reaction. The guidance encourages organisations to act now, before frontier AI systems become deeply embedded into operational environments without appropriate governance, oversight, and cybersecurity controls. The challenge is not simply securing AI itself but rather the challenge is securing what AI identities can access, influence, control, and automate.
Historically, cybersecurity focused on protecting networks, endpoints, servers, applications, SaaS, and even the cloud. Today, threat actors are increasingly targeting identities because identities with excessive privileges provide the proverbial keys to the kingdom. AI accelerates this challenge dramatically due to all connections an AI agent can make and operate as the new middleware layer within an organisation.
Modern AI systems often require access to databases, APIs, cloud resources, SaaS applications, and sensitive business information and they frequently operate using service accounts, API keys, machine credentials, and privileged permissions. In many environments, these AI-powered workloads already possess more access than individual employees and represent a massive risk surface of standing privileges.
With this in mind, the PSPF advisory implicitly highlights a critical question every organisation should ask, “who is governing the identities behind AI?”. The answer cannot be traditional identity and access management solutions and teams alone. It requires comprehensive identity security strategy that spans human identities, machine identities, service accounts, cloud entitlements, and emerging AI agents (and corresponding agentic AI systems).
Organisations cannot secure what they cannot see, and as AI adoption accelerates, enterprises are creating thousands of new non-human identities through automation, integrations, service accounts, APIs, robotic process automation, and agentic AI frameworks. Many organisations lack complete visibility into these identities and the entitlements (privileges, permissions, and rights) they possess. Research consistently shows that non-human identities are growing faster than human identities, creating significant governance gaps.
The good news is that Australian enterprises can address this challenge by discovering privileged identities, mapping attack paths, identifying excessive permissions, and exposing hidden risks across hybrid environments. Instead of relying on static inventories and time-based governance solutions, organisations gain continuous visibility into who or what has access to critical resources and if any malicious behaviour is actually occurring.
AI systems operate at machine speed. It is a cliche we all must accept because of the speed something can go awry. A compromised AI agent with excessive permissions can access sensitive information, modify infrastructure, disable controls, or perform actions at a scale far beyond human capabilities especially if compromised by a threat actor.
The advisory’s focus on cyber readiness requires organisations to eliminate unnecessary standing privileges before these risks materialise. To address these concerns, organisations can deploy solutions which enable least privilege enforcement, credential and secrets management, session monitoring, and Just-in-Time (JIT) access. Rather than granting persistent privileges (administrators and technically any other identities), access can be provisioned only when required and revoked immediately afterward based on mission, task, and change control requirements. This dramatically reduces the attack surface available to both human adversaries and compromised AI systems.
However, some of the most difficult questions surrounding frontier AI revolve around determining responsibility, accountability, and ownership. When an AI agent performs an action, accesses data, or initiates a workflow, organisations must be able to answer who authorised the action, what permissions were used, and whether the activity was appropriate. The PSPF advisory emphasises readiness, which inherently requires auditability and human oversight.
Finally, the advisory reinforces principles that align closely with Zero Trust Architectures Continuous verification and monitoring, policy-based access, segmentation, and identity-centric security are foundational requirements for operating securely in an AI-driven environment. The frontier AI policy also does not eliminate existing cybersecurity principles but rather amplifies their importance as AI is integrated into existing environments and legacy solutions. Disciplines contained within zero trust and accepted widely as best practices like least privilege, privileged access management, identity governance, and continuous monitoring become even more critical when intelligent systems operate alongside human users and existing infrastructure.
The central message of PSPF Policy Advisory 001-2026 is crystal clear. Australian enterprises must prepare now for a future where AI acts with increasing autonomy and the most effective way to secure that future is not by focusing solely on AI models and generative output. It is by securing the identities, privileges, and access paths that enable AI to operate as a true partner within an organisation.
As we enter the new financial year, CISOs should consider privileged centric identity security the control plane for AI governance which in turn will enable them to operationalise that strategy while maintaining visibility, intelligence, and protection in the frontier AI era.
