If you wanted to get inside a secure compound, the movies would tell you to cut the fence or scale the wall. In real life, that's not how it works. Walls get defended. Gates get watched. What actually gets you in is a uniform, a clipboard, and a confident nod to the guard. No alarm. No log. By the time anyone notices you were there, you've already left with what you came for.
That's shadow AI. It didn't break in. Somebody waved it through the gate because it looked useful and harmless. Nobody logged it, and now it's standing inside your business, touching data nobody signed off on.
Raj Kunjira, who heads up GRC, AI security and vCISO at StickmanCyber, opened a recent live session with one question: which data has an AI agent already touched in your organisation?
For most businesses, the honest answer is “we don't know.” An agent somewhere already has access to an application, a system, maybe sensitive data, and nobody approved it or logged it. Individually, none of this looks dangerous. A salesperson turns on an AI note-taker instead of visibly recording a meeting. A copilot gets wired into email. Someone builds a weekend automation with a personal API key. Each one is harmless on its own. Together, they're an unmanaged surface where company data is leaving the building without anyone deciding it should.
Four places cover most of it, and none of them show up on a standard asset register:
|
1. Browser extensions that read every page and form your team opens, including meeting tools like Read.ai |
|
2. Copilots wired into email, already sitting inside Outlook or Google Workspace, with access to inboxes and calendars and the ability to act on someone's behalf |
|
3. Personal API keys, where individuals pipe company data straight into AI services outside any contract. |
|
4. Weekend automations, built by capable people trying to save time, that never went through review |
IBM's 2025 Cost of a Data Breach Report puts a number on this. A breach involving shadow AI costs an average of $6.43 million AUD. Without it, that figure drops to $5.5 million, a $930,000 gap. In the same survey, 97% of organisations that suffered an AI-related incident had no proper AI access controls, and 63% had no AI governance policy at all.
This isn't a technology failure. It's a governance gap, and governance gaps are things you can actually close.
When StickmanCyber runs a dashboard review of an organisation's AI usage, three categories of data keep surfacing that leaders were confident were untouched: personal information pasted into tools, intellectual property flowing to third-party AI services, and financial records sitting in places nobody tracked. The problem isn't that AI is dangerous or that people are careless. It's that nobody could answer the simple question of what it had already touched.
Telling a board “we have unmanaged AI” gets a shrug. Telling them “we have sensitive data flowing to three unvetted vendors, which breaches clauses in our largest contracts and may void our cyber insurance” gets a budget. Sizing shadow AI in business terms means looking at three things: what data is actually leaving the building, which third parties are processing it and under whose terms, and what you've already promised clients, regulators and insurers about how that data gets handled.
Boards are also being asked to prove this, not just assure it. The Australian Institute of Company Directors released a governance framework in June 2026 built around four questions: Does AI use fit your risk appetite (strategy)? Who owns AI oversight (structure)? Does your policy actually restrict unapproved tools with access to sensitive data (practices)? And is AI literacy in place for staff and contractors (enablers)? Directors don't need to be AI experts, but they do need a live AI register, a named accountable owner, and a policy with real boundaries. Assurance won't cut it anymore. Evidence will.
The instinct to ban everything is the wrong one. Ban shadow AI outright and people don't stop using it, they just stop telling you. The exposure gets worse, not better. The plan that actually works moves in four stages:
Week 1, Discovery: inventory every tool, copilot, API key and automation across the four places above. The goal is visibility, not enforcement.
Week 2, Triage: rank what you found by data and contract exposure, so you focus on what actually matters first.
Week 3, Sanction: approve safe, supported alternatives people will genuinely use. Without a good option, they go back to shadow.
Week 4, Guardrails: put access controls, policy and monitoring in place.
Lead with visibility and sanctioned alternatives, and the business moves with you instead of around you.
Shadow AI already exists in your organisation. It's not a question of if, it's whether you can see it. It's a governance gap, not a technology failure, which means it's fixable and it sits with leadership, not IT. And the fix starts with the 30-day plan above, not a freeze.
Answer the question “which data has AI already touched in my organisation” on your own terms now, rather than explaining the gap to a customer, a board, or a regulator later.
Want the full picture? This post covers the framework. The live session goes deeper, including a Q&A on who should own AI accountability, whether SME businesses need formal governance, and how shadow AI intersects with the Notifiable Data Breaches scheme. Watch the full webinar recording →
Not sure where your business stands on shadow AI? Book a free consultation with StickmanCyber's team and get a clear view of what's already running inside your organisation, mapped to your own stack.