Australian Agencies Face New Zero-Trust Priorities for AI Agents

An October 2026 Fortinet opinion article by Cornelius Mare, Chief Information Security Officer for Australia at Fortinet, examines how agentic artificial intelligence is changing identity and access security. Earlier discussions of generative AI often centered on employee use: which tools staff could access, what information they could enter, and how to prevent sensitive data from leaving controlled environments. That concern remains, but AI agents introduce a different challenge: the entity requesting access to an application, dataset, or service may no longer be a person. Agents can perform tasks, call tools, interact with applications and data, and communicate with other agents, becoming active participants in business processes rather than simply tools used by employees.

The article argues that this development strains security assumptions built around people and devices. Traditional zero-trust approaches establish who a user is, which device they use, and what access should be permitted under specific conditions. Agentic systems can place several machine-to-machine interactions behind a single human request. For example, an agent may retrieve a customer record, query another system, analyse information, generate an output, and initiate a follow-on action. Security teams therefore need to assess not only whether an identity is authenticated, but also what it is doing, which resources it is trying to reach, and whether its activity fits its assigned task.

The procurement relevance is practical rather than tied to a specific acquisition. Australian agencies and contractors evaluating AI-enabled systems can use the article’s recommendations to shape security requirements, supplier questionnaires, architecture reviews, and operational acceptance criteria. Contractors serving the Australian Government may be better positioned to explain how their products identify agents, limit permissions to defined tasks, monitor activity, and preserve logs under each agent’s own identity. These are planning considerations drawn from the article; it does not establish a government mandate, announce a solicitation, identify a contract value, or provide a procurement contact or implementation deadline.

The article connects its analysis to guidance released in May 2026 by the Australian Signals Directorate’s Australian Cyber Security Centre (ACSC) and international partners on the careful adoption of agentic AI services. It says that guidance recommends strict privilege management, continuous monitoring, strong identity management, and human oversight. For government buyers, the implications include asking suppliers to demonstrate how controls apply across agent-to-agent and agent-to-system activity, rather than relying solely on conventional user authentication. The source presents these priorities as relevant to cybersecurity planning and supplier evaluation, not as a new binding procurement rule.

• Agentic AI expands the zero-trust scope from human users and devices to non-human identities and autonomous agents operating across systems.
• May 2026 ACSC guidance, issued with international partners, is described as recommending strong identity management, strict privilege controls, continuous monitoring, and human oversight.
• Inventory existing agents across environments so security teams can identify what is operating, where it connects, and which business processes it supports.
• Assign each agent task-scoped credentials rather than allowing it to inherit a user’s broader permissions.
• Record activity under the agent’s own identity to support visibility, investigation, and accountability for actions taken across applications and services.
• Assess agent access across an entire workflow: one human request may trigger multiple machine-to-machine transactions and interactions with separate resources.
• Suppliers can document how their systems authenticate agents, restrict permissions, monitor behavior, and retain attributable activity logs for government customers.
• The article offers security and procurement-planning considerations, but specifies no solicitation, contract value, procurement contact, government mandate, or implementation deadline.