Half of ANZ Firms Run AI Agents Without Governance

A new wave of research is sounding the alarm on a dangerous mismatch in enterprise security: organisations across Australia and New Zealand are racing to deploy autonomous AI agents while leaving governance frameworks behind. According to the findings, roughly half of ANZ organisations are running AI agents without formal oversight — and this is happening precisely as deepfake-driven fraud, impersonation, and social engineering attacks escalate across the region.

The convergence of these two trends — ungoverned agentic AI and increasingly convincing synthetic media — represents one of the most underappreciated risk vectors facing enterprises today. For an audience focused on digital authenticity, this is a critical signal about how the attack surface is expanding faster than defensive controls.

The Governance Gap

AI agents differ fundamentally from the chatbots and copilots that defined the first wave of generative AI adoption. Rather than simply answering prompts, agents can take actions: querying databases, sending communications, executing transactions, and chaining together multi-step workflows with limited human supervision. That autonomy is exactly what makes them productive — and exactly what makes ungoverned deployment hazardous.

When the research finds that half of ANZ organisations operate these systems without governance, it points to a lack of basic controls: no clear policies on what data agents can access, no audit trails of agent decisions, no identity verification layers, and no guardrails against manipulation. In practice, an unmonitored agent with access to financial systems or sensitive communications is a high-value target.

Why Deepfakes Amplify the Risk

The deepfake angle is where this story becomes especially relevant to the synthetic media landscape. Voice cloning and video deepfakes have already been used in high-profile fraud cases — including the widely reported incident in which an employee was tricked into transferring millions after a video call populated entirely by deepfaked executives.

Combine that with autonomous AI agents and the threat compounds. Consider an agent designed to act on instructions received via email or voice. If an attacker can clone an executive's voice or spoof a video identity convincingly, an ungoverned agent may execute fraudulent instructions without the friction a human reviewer might apply. The same synthetic media techniques used to deceive people can now be deployed against automated systems that lack identity verification and provenance checks.

This is the core authenticity problem: as both content generation and content consumption become increasingly automated, the ability to verify who is actually issuing a request — and whether the media accompanying it is genuine — becomes mission-critical.

What Effective Governance Looks Like

Closing the gap requires more than policy documents. Security leaders recommend a layered approach that combines technical and procedural controls:

• Identity and access management for agents: Treating AI agents as first-class identities with scoped permissions, rather than inheriting broad human credentials.
• Content authentication: Deploying deepfake detection and provenance verification (such as content credentials and watermarking standards) at the points where agents ingest media or instructions.
• Human-in-the-loop checkpoints: Requiring verification for high-risk actions like payments or data exports, regardless of how authentic a request appears.
• Audit logging: Maintaining tamper-evident records of every agent action to enable forensic review.
• Out-of-band verification: Confirming sensitive requests through a separate, trusted channel — a simple but powerful defence against voice and video impersonation.

A Regional Signal With Global Implications

While the research focuses on ANZ, the pattern is far from regional. Enterprises worldwide are under pressure to adopt agentic AI to stay competitive, and governance frameworks consistently lag behind deployment velocity. The ANZ data simply quantifies a trend that security teams everywhere are confronting.

For organisations building defensive postures, the takeaway is clear: AI governance and synthetic media detection can no longer be treated as separate workstreams. As autonomous systems take on more decision-making authority, the integrity of the inputs they act upon — whether text, voice, or video — becomes a primary security concern. Verifying authenticity at machine speed will be one of the defining challenges of the agentic era.

The half of ANZ organisations that have governance in place are better positioned not just for compliance, but for resilience against an emerging class of threats where synthetic media and autonomous action intersect. The other half are, quite literally, running on trust they can no longer afford to assume.

Stay informed on AI video and digital authenticity. Follow Skrew AI News.