AI Governance Becomes a Business Requirement: What OneTrust's 2027 Predictions Mean for Australian Organisations
OneTrust has released a 2027 predictions report arguing that Australian organisations will face rising commercial and regulatory pressure to show their AI systems are safe, controlled and accountable. The report, Architecting AI Governance: Blueprints for a Moving Maze, draws on input from more than 20 global experts and describes four developments it expects to shape AI adoption and trust.
The company points to signs that governance is lagging behind adoption. In its 2026 AI-Ready Governance Report, 87% of respondents said their organisations encourage AI agent use, but only 47% said they have clear governance, oversight and controls in place. Another 40% said those capabilities are still developing. In Australia, 48% of organisations reported limited visibility into employee-led or unsanctioned AI use, and 52% said concerns about data quality, access, privacy or security had delayed launching or scaling AI projects.
The four predictions are: AI assurance becomes a commercial gate, with buyers wanting current evidence before enabling or expanding AI capabilities; an approved AI agent can still take unauthorised action, because technical access does not equal delegated authority; material changes to an AI system's capabilities, permissions or operating conditions will trigger fresh assurance reviews; and customer data-use choices will need to carry through into later AI processing, rather than consent being treated as a blanket approval.
Blair Hasforth, OneTrust's Regional Vice President Sales, APAC, said an employee's access to a system does not automatically mean an AI agent should have the same authority. He said businesses need to be deliberate about the boundaries and oversight they set. The report also argues that adding headcount or relying on static policies will not be enough, and calls for governance models that adapt as technology and use cases change.
Why it matters
Many businesses grant AI agents the same system access as the employee using them, but the report argues access is not authority, so permission models built for people may leave agents able to act beyond what anyone intended. Limited visibility into unsanctioned AI use (48% in Australia) means some exposure sits outside any approval process. Customers may also soon ask for current evidence of AI assurance, so one-off approvals and static policies become a weak point.
What to do
- Inventory the AI tools and agents in use, including unsanctioned employee use.
- Define and record what each AI agent is authorised to do, separate from the access it technically has.
- Reassess approvals whenever an AI system's capabilities, permissions or operating conditions change.
- Check that customer data-use choices are respected in any AI processing of that data.