Across Australian organisations, teams are standing up their own AI tools — chat assistants, model servers, knowledge bases, and automation builders — faster than security can keep track of them. Many end up quietly exposed to the internet. TONK, DGplex's external attack surface management platform, now finds them.
The Challenge: Shadow AI on the Perimeter
The rush to adopt AI has a security side effect that few organisations have caught up with. Where sensitive workloads once meant a database or an admin console, they now include self-hosted AI systems — the servers that run models, the databases that store the knowledge those models draw on, and the tools that wire them into business processes.
These systems are often spun up by developers or business units in a hurry, outside the usual provisioning process, and pushed live without authentication. An exposed AI system is not a niche problem: it can leak the private data an organisation fed into it, expose the internal knowledge base behind a chat assistant, or hand an attacker a foothold to run their own workloads on your infrastructure. Because these tools are so new, most attack surface tools don't recognise them at all — so they never appear on a security team's radar until something goes wrong.
You can't defend what you don't know is exposed — and right now, AI infrastructure is the fastest-growing blind spot on the corporate perimeter.
What's New: AI Exposure Discovery
TONK now actively looks for exposed AI infrastructure across your external attack surface, alongside the vulnerabilities, misconfigurations, and exposed services it already surfaces. If one of your internet-facing hosts is running an AI system that anyone can reach without logging in, you'll see it — with the same clarity, severity rating, and triage workflow as any other finding.
It works the way the rest of the platform does: continuously, from the outside, exactly as an attacker would see you. There's nothing to install and no agent to deploy. New exposures surface automatically as your estate changes, so a system that goes live on a Friday afternoon doesn't wait until the next audit to be noticed.
Just as importantly, it extends to the loose ends that AI adoption leaves behind — the configuration and credential files that AI tooling can inadvertently publish to a web server. These are the modern equivalent of a leaked API key, and they're just as dangerous. Our checks confirm that an exposure is real without ever reading the sensitive data behind it.
What You Can Now See
- Exposed AI systems — self-hosted AI services reachable from the internet without authentication, surfaced the moment they appear.
- Leaked AI configuration and credentials — sensitive AI-tooling files inadvertently published to a public web server.
- One prioritised view — AI exposures sit alongside your vulnerabilities and misconfigurations, ranked by real-world impact, not treated as a separate silo.
- Safe by design — every check confirms the exposure without reading or retrieving the data behind it.
Who This Is For
This matters most for organisations moving quickly on AI — which, in 2026, is nearly everyone. Security leaders subject to the Notifiable Data Breaches scheme, APRA CPS 234, or the Protective Security Policy Framework (PSPF) have a particular interest: an exposed AI knowledge base can hold exactly the kind of data whose disclosure triggers a regulatory obligation. Giving your security team visibility of it before an attacker finds it is the difference between a quiet fix and a public incident.
How It Fits
AI exposure discovery is part of TONK's continuous external attack surface management — the same platform that maps your internet-facing assets, verifies real vulnerabilities, and tracks them through to remediation. It's built in Australia, hosted in Australia, and operated by Australians, so your attack-surface data never leaves Australian jurisdiction.